fix(ci): update Go version check in release workflow to 1.26.1

fix(openai): 统一官方 Codex 客户端识别逻辑

.github/workflows/backend-ci.yml

@@ -19,7 +19,7 @@ jobs:
           cache: true
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: Unit tests
         working-directory: backend
         run: make test-unit
@@ -38,10 +38,10 @@ jobs:
           cache: true
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: golangci-lint
         uses: golangci/golangci-lint-action@v9
         with:
-          version: v2.7
+          version: v2.11
           args: --timeout=30m
           working-directory: backend

.github/workflows/release.yml

@@ -115,7 +115,7 @@ jobs:
 
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
 
       # Docker setup for GoReleaser
       - name: Set up QEMU

.github/workflows/security-scan.yml

@@ -23,7 +23,7 @@ jobs:
           cache-dependency-path: backend/go.sum
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: Run govulncheck
         working-directory: backend
         run: |

Dockerfile

@@ -7,7 +7,7 @@
 # =============================================================================
 
 ARG NODE_IMAGE=node:24-alpine
-ARG GOLANG_IMAGE=golang:1.25.7-alpine
+ARG GOLANG_IMAGE=golang:1.26.1-alpine
 ARG ALPINE_IMAGE=alpine:3.21
 ARG GOPROXY=https://goproxy.cn,direct
 ARG GOSUMDB=sum.golang.google.cn

backend/.golangci.yml

@@ -93,20 +93,13 @@ linters:
       check-escaping-errors: true
     staticcheck:
       # https://staticcheck.dev/docs/configuration/options/#dot_import_whitelist
-      # Default: ["github.com/mmcloughlin/avo/build", "github.com/mmcloughlin/avo/operand", "github.com/mmcloughlin/avo/reg"]
       dot-import-whitelist:
         - fmt
       # https://staticcheck.dev/docs/configuration/options/#initialisms
-      # Default: ["ACL", "API", "ASCII", "CPU", "CSS", "DNS", "EOF", "GUID", "HTML", "HTTP", "HTTPS", "ID", "IP", "JSON", "QPS", "RAM", "RPC", "SLA", "SMTP", "SQL", "SSH", "TCP", "TLS", "TTL", "UDP", "UI", "GID", "UID", "UUID", "URI", "URL", "UTF8", "VM", "XML", "XMPP", "XSRF", "XSS", "SIP", "RTP", "AMQP", "DB", "TS"]
       initialisms: [ "ACL", "API", "ASCII", "CPU", "CSS", "DNS", "EOF", "GUID", "HTML", "HTTP", "HTTPS", "ID", "IP", "JSON", "QPS", "RAM", "RPC", "SLA", "SMTP", "SQL", "SSH", "TCP", "TLS", "TTL", "UDP", "UI", "GID", "UID", "UUID", "URI", "URL", "UTF8", "VM", "XML", "XMPP", "XSRF", "XSS", "SIP", "RTP", "AMQP", "DB", "TS" ]
       # https://staticcheck.dev/docs/configuration/options/#http_status_code_whitelist
-      # Default: ["200", "400", "404", "500"]
       http-status-code-whitelist: [ "200", "400", "404", "500" ]
-      # SAxxxx checks in https://staticcheck.dev/docs/configuration/options/#checks
-      # Example (to disable some checks): [ "all", "-SA1000", "-SA1001"]
-      # Run `GL_DEBUG=staticcheck golangci-lint run --enable=staticcheck` to see all available checks and enabled by config checks.
-      # Default: ["all", "-ST1000", "-ST1003", "-ST1016", "-ST1020", "-ST1021", "-ST1022"]
-      # Temporarily disable style checks to allow CI to pass
+      # "all" enables every SA/ST/S/QF check; only list the ones to disable.
       checks:
         - all
         - -ST1000  # Package comment format
@@ -114,489 +107,19 @@ linters:
         - -ST1020  # Comment on exported method format
         - -ST1021  # Comment on exported type format
         - -ST1022  # Comment on exported variable format
-        # Invalid regular expression.
-        # https://staticcheck.dev/docs/checks/#SA1000
-        - SA1000
-        # Invalid template.
-        # https://staticcheck.dev/docs/checks/#SA1001
-        - SA1001
-        # Invalid format in 'time.Parse'.
-        # https://staticcheck.dev/docs/checks/#SA1002
-        - SA1002
-        # Unsupported argument to functions in 'encoding/binary'.
-        # https://staticcheck.dev/docs/checks/#SA1003
-        - SA1003
-        # Suspiciously small untyped constant in 'time.Sleep'.
-        # https://staticcheck.dev/docs/checks/#SA1004
-        - SA1004
-        # Invalid first argument to 'exec.Command'.
-        # https://staticcheck.dev/docs/checks/#SA1005
-        - SA1005
-        # 'Printf' with dynamic first argument and no further arguments.
-        # https://staticcheck.dev/docs/checks/#SA1006
-        - SA1006
-        # Invalid URL in 'net/url.Parse'.
-        # https://staticcheck.dev/docs/checks/#SA1007
-        - SA1007
-        # Non-canonical key in 'http.Header' map.
-        # https://staticcheck.dev/docs/checks/#SA1008
-        - SA1008
-        # '(*regexp.Regexp).FindAll' called with 'n == 0', which will always return zero results.
-        # https://staticcheck.dev/docs/checks/#SA1010
-        - SA1010
-        # Various methods in the "strings" package expect valid UTF-8, but invalid input is provided.
-        # https://staticcheck.dev/docs/checks/#SA1011
-        - SA1011
-        # A nil 'context.Context' is being passed to a function, consider using 'context.TODO' instead.
-        # https://staticcheck.dev/docs/checks/#SA1012
-        - SA1012
-        # 'io.Seeker.Seek' is being called with the whence constant as the first argument, but it should be the second.
-        # https://staticcheck.dev/docs/checks/#SA1013
-        - SA1013
-        # Non-pointer value passed to 'Unmarshal' or 'Decode'.
-        # https://staticcheck.dev/docs/checks/#SA1014
-        - SA1014
-        # Using 'time.Tick' in a way that will leak. Consider using 'time.NewTicker', and only use 'time.Tick' in tests, commands and endless functions.
-        # https://staticcheck.dev/docs/checks/#SA1015
-        - SA1015
-        # Trapping a signal that cannot be trapped.
-        # https://staticcheck.dev/docs/checks/#SA1016
-        - SA1016
-        # Channels used with 'os/signal.Notify' should be buffered.
-        # https://staticcheck.dev/docs/checks/#SA1017
-        - SA1017
-        # 'strings.Replace' called with 'n == 0', which does nothing.
-        # https://staticcheck.dev/docs/checks/#SA1018
-        - SA1018
-        # Using a deprecated function, variable, constant or field.
-        # https://staticcheck.dev/docs/checks/#SA1019
-        - SA1019
-        # Using an invalid host:port pair with a 'net.Listen'-related function.
-        # https://staticcheck.dev/docs/checks/#SA1020
-        - SA1020
-        # Using 'bytes.Equal' to compare two 'net.IP'.
-        # https://staticcheck.dev/docs/checks/#SA1021
-        - SA1021
-        # Modifying the buffer in an 'io.Writer' implementation.
-        # https://staticcheck.dev/docs/checks/#SA1023
-        - SA1023
-        # A string cutset contains duplicate characters.
-        # https://staticcheck.dev/docs/checks/#SA1024
-        - SA1024
-        # It is not possible to use '(*time.Timer).Reset''s return value correctly.
-        # https://staticcheck.dev/docs/checks/#SA1025
-        - SA1025
-        # Cannot marshal channels or functions.
-        # https://staticcheck.dev/docs/checks/#SA1026
-        - SA1026
-        # Atomic access to 64-bit variable must be 64-bit aligned.
-        # https://staticcheck.dev/docs/checks/#SA1027
-        - SA1027
-        # 'sort.Slice' can only be used on slices.
-        # https://staticcheck.dev/docs/checks/#SA1028
-        - SA1028
-        # Inappropriate key in call to 'context.WithValue'.
-        # https://staticcheck.dev/docs/checks/#SA1029
-        - SA1029
-        # Invalid argument in call to a 'strconv' function.
-        # https://staticcheck.dev/docs/checks/#SA1030
-        - SA1030
-        # Overlapping byte slices passed to an encoder.
-        # https://staticcheck.dev/docs/checks/#SA1031
-        - SA1031
-        # Wrong order of arguments to 'errors.Is'.
-        # https://staticcheck.dev/docs/checks/#SA1032
-        - SA1032
-        # 'sync.WaitGroup.Add' called inside the goroutine, leading to a race condition.
-        # https://staticcheck.dev/docs/checks/#SA2000
-        - SA2000
-        # Empty critical section, did you mean to defer the unlock?.
-        # https://staticcheck.dev/docs/checks/#SA2001
-        - SA2001
-        # Called 'testing.T.FailNow' or 'SkipNow' in a goroutine, which isn't allowed.
-        # https://staticcheck.dev/docs/checks/#SA2002
-        - SA2002
-        # Deferred 'Lock' right after locking, likely meant to defer 'Unlock' instead.
-        # https://staticcheck.dev/docs/checks/#SA2003
-        - SA2003
-        # 'TestMain' doesn't call 'os.Exit', hiding test failures.
-        # https://staticcheck.dev/docs/checks/#SA3000
-        - SA3000
-        # Assigning to 'b.N' in benchmarks distorts the results.
-        # https://staticcheck.dev/docs/checks/#SA3001
-        - SA3001
-        # Binary operator has identical expressions on both sides.
-        # https://staticcheck.dev/docs/checks/#SA4000
-        - SA4000
-        # '&*x' gets simplified to 'x', it does not copy 'x'.
-        # https://staticcheck.dev/docs/checks/#SA4001
-        - SA4001
-        # Comparing unsigned values against negative values is pointless.
-        # https://staticcheck.dev/docs/checks/#SA4003
-        - SA4003
-        # The loop exits unconditionally after one iteration.
-        # https://staticcheck.dev/docs/checks/#SA4004
-        - SA4004
-        # Field assignment that will never be observed. Did you mean to use a pointer receiver?.
-        # https://staticcheck.dev/docs/checks/#SA4005
-        - SA4005
-        # A value assigned to a variable is never read before being overwritten. Forgotten error check or dead code?.
-        # https://staticcheck.dev/docs/checks/#SA4006
-        - SA4006
-        # The variable in the loop condition never changes, are you incrementing the wrong variable?.
-        # https://staticcheck.dev/docs/checks/#SA4008
-        - SA4008
-        # A function argument is overwritten before its first use.
-        # https://staticcheck.dev/docs/checks/#SA4009
-        - SA4009
-        # The result of 'append' will never be observed anywhere.
-        # https://staticcheck.dev/docs/checks/#SA4010
-        - SA4010
-        # Break statement with no effect. Did you mean to break out of an outer loop?.
-        # https://staticcheck.dev/docs/checks/#SA4011
-        - SA4011
-        # Comparing a value against NaN even though no value is equal to NaN.
-        # https://staticcheck.dev/docs/checks/#SA4012
-        - SA4012
-        # Negating a boolean twice ('!!b') is the same as writing 'b'. This is either redundant, or a typo.
-        # https://staticcheck.dev/docs/checks/#SA4013
-        - SA4013
-        # An if/else if chain has repeated conditions and no side-effects; if the condition didn't match the first time, it won't match the second time, either.
-        # https://staticcheck.dev/docs/checks/#SA4014
-        - SA4014
-        # Calling functions like 'math.Ceil' on floats converted from integers doesn't do anything useful.
-        # https://staticcheck.dev/docs/checks/#SA4015
-        - SA4015
-        # Certain bitwise operations, such as 'x ^ 0', do not do anything useful.
-        # https://staticcheck.dev/docs/checks/#SA4016
-        - SA4016
-        # Discarding the return values of a function without side effects, making the call pointless.
-        # https://staticcheck.dev/docs/checks/#SA4017
-        - SA4017
-        # Self-assignment of variables.
-        # https://staticcheck.dev/docs/checks/#SA4018
-        - SA4018
-        # Multiple, identical build constraints in the same file.
-        # https://staticcheck.dev/docs/checks/#SA4019
-        - SA4019
-        # Unreachable case clause in a type switch.
-        # https://staticcheck.dev/docs/checks/#SA4020
-        - SA4020
-        # "x = append(y)" is equivalent to "x = y".
-        # https://staticcheck.dev/docs/checks/#SA4021
-        - SA4021
-        # Comparing the address of a variable against nil.
-        # https://staticcheck.dev/docs/checks/#SA4022
-        - SA4022
-        # Impossible comparison of interface value with untyped nil.
-        # https://staticcheck.dev/docs/checks/#SA4023
-        - SA4023
-        # Checking for impossible return value from a builtin function.
-        # https://staticcheck.dev/docs/checks/#SA4024
-        - SA4024
-        # Integer division of literals that results in zero.
-        # https://staticcheck.dev/docs/checks/#SA4025
-        - SA4025
-        # Go constants cannot express negative zero.
-        # https://staticcheck.dev/docs/checks/#SA4026
-        - SA4026
-        # '(*net/url.URL).Query' returns a copy, modifying it doesn't change the URL.
-        # https://staticcheck.dev/docs/checks/#SA4027
-        - SA4027
-        # 'x % 1' is always zero.
-        # https://staticcheck.dev/docs/checks/#SA4028
-        - SA4028
-        # Ineffective attempt at sorting slice.
-        # https://staticcheck.dev/docs/checks/#SA4029
-        - SA4029
-        # Ineffective attempt at generating random number.
-        # https://staticcheck.dev/docs/checks/#SA4030
-        - SA4030
-        # Checking never-nil value against nil.
-        # https://staticcheck.dev/docs/checks/#SA4031
-        - SA4031
-        # Comparing 'runtime.GOOS' or 'runtime.GOARCH' against impossible value.
-        # https://staticcheck.dev/docs/checks/#SA4032
-        - SA4032
-        # Assignment to nil map.
-        # https://staticcheck.dev/docs/checks/#SA5000
-        - SA5000
-        # Deferring 'Close' before checking for a possible error.
-        # https://staticcheck.dev/docs/checks/#SA5001
-        - SA5001
-        # The empty for loop ("for {}") spins and can block the scheduler.
-        # https://staticcheck.dev/docs/checks/#SA5002
-        - SA5002
-        # Defers in infinite loops will never execute.
-        # https://staticcheck.dev/docs/checks/#SA5003
-        - SA5003
-        # "for { select { ..." with an empty default branch spins.
-        # https://staticcheck.dev/docs/checks/#SA5004
-        - SA5004
-        # The finalizer references the finalized object, preventing garbage collection.
-        # https://staticcheck.dev/docs/checks/#SA5005
-        - SA5005
-        # Infinite recursive call.
-        # https://staticcheck.dev/docs/checks/#SA5007
-        - SA5007
-        # Invalid struct tag.
-        # https://staticcheck.dev/docs/checks/#SA5008
-        - SA5008
-        # Invalid Printf call.
-        # https://staticcheck.dev/docs/checks/#SA5009
-        - SA5009
-        # Impossible type assertion.
-        # https://staticcheck.dev/docs/checks/#SA5010
-        - SA5010
-        # Possible nil pointer dereference.
-        # https://staticcheck.dev/docs/checks/#SA5011
-        - SA5011
-        # Passing odd-sized slice to function expecting even size.
-        # https://staticcheck.dev/docs/checks/#SA5012
-        - SA5012
-        # Using 'regexp.Match' or related in a loop, should use 'regexp.Compile'.
-        # https://staticcheck.dev/docs/checks/#SA6000
-        - SA6000
-        # Missing an optimization opportunity when indexing maps by byte slices.
-        # https://staticcheck.dev/docs/checks/#SA6001
-        - SA6001
-        # Storing non-pointer values in 'sync.Pool' allocates memory.
-        # https://staticcheck.dev/docs/checks/#SA6002
-        - SA6002
-        # Converting a string to a slice of runes before ranging over it.
-        # https://staticcheck.dev/docs/checks/#SA6003
-        - SA6003
-        # Inefficient string comparison with 'strings.ToLower' or 'strings.ToUpper'.
-        # https://staticcheck.dev/docs/checks/#SA6005
-        - SA6005
-        # Using io.WriteString to write '[]byte'.
-        # https://staticcheck.dev/docs/checks/#SA6006
-        - SA6006
-        # Defers in range loops may not run when you expect them to.
-        # https://staticcheck.dev/docs/checks/#SA9001
-        - SA9001
-        # Using a non-octal 'os.FileMode' that looks like it was meant to be in octal.
-        # https://staticcheck.dev/docs/checks/#SA9002
-        - SA9002
-        # Empty body in an if or else branch.
-        # https://staticcheck.dev/docs/checks/#SA9003
-        - SA9003
-        # Only the first constant has an explicit type.
-        # https://staticcheck.dev/docs/checks/#SA9004
-        - SA9004
-        # Trying to marshal a struct with no public fields nor custom marshaling.
-        # https://staticcheck.dev/docs/checks/#SA9005
-        - SA9005
-        # Dubious bit shifting of a fixed size integer value.
-        # https://staticcheck.dev/docs/checks/#SA9006
-        - SA9006
-        # Deleting a directory that shouldn't be deleted.
-        # https://staticcheck.dev/docs/checks/#SA9007
-        - SA9007
-        # 'else' branch of a type assertion is probably not reading the right value.
-        # https://staticcheck.dev/docs/checks/#SA9008
-        - SA9008
-        # Ineffectual Go compiler directive.
-        # https://staticcheck.dev/docs/checks/#SA9009
-        - SA9009
-        # NOTE: ST1000, ST1001, ST1003, ST1020, ST1021, ST1022 are disabled above
-        # Incorrectly formatted error string.
-        # https://staticcheck.dev/docs/checks/#ST1005
-        - ST1005
-        # Poorly chosen receiver name.
-        # https://staticcheck.dev/docs/checks/#ST1006
-        - ST1006
-        # A function's error value should be its last return value.
-        # https://staticcheck.dev/docs/checks/#ST1008
-        - ST1008
-        # Poorly chosen name for variable of type 'time.Duration'.
-        # https://staticcheck.dev/docs/checks/#ST1011
-        - ST1011
-        # Poorly chosen name for error variable.
-        # https://staticcheck.dev/docs/checks/#ST1012
-        - ST1012
-        # Should use constants for HTTP error codes, not magic numbers.
-        # https://staticcheck.dev/docs/checks/#ST1013
-        - ST1013
-        # A switch's default case should be the first or last case.
-        # https://staticcheck.dev/docs/checks/#ST1015
-        - ST1015
-        # Use consistent method receiver names.
-        # https://staticcheck.dev/docs/checks/#ST1016
-        - ST1016
-        # Don't use Yoda conditions.
-        # https://staticcheck.dev/docs/checks/#ST1017
-        - ST1017
-        # Avoid zero-width and control characters in string literals.
-        # https://staticcheck.dev/docs/checks/#ST1018
-        - ST1018
-        # Importing the same package multiple times.
-        # https://staticcheck.dev/docs/checks/#ST1019
-        - ST1019
-        # NOTE: ST1020, ST1021, ST1022 removed (disabled above)
-        # Redundant type in variable declaration.
-        # https://staticcheck.dev/docs/checks/#ST1023
-        - ST1023
-        # Use plain channel send or receive instead of single-case select.
-        # https://staticcheck.dev/docs/checks/#S1000
-        - S1000
-        # Replace for loop with call to copy.
-        # https://staticcheck.dev/docs/checks/#S1001
-        - S1001
-        # Omit comparison with boolean constant.
-        # https://staticcheck.dev/docs/checks/#S1002
-        - S1002
-        # Replace call to 'strings.Index' with 'strings.Contains'.
-        # https://staticcheck.dev/docs/checks/#S1003
-        - S1003
-        # Replace call to 'bytes.Compare' with 'bytes.Equal'.
-        # https://staticcheck.dev/docs/checks/#S1004
-        - S1004
-        # Drop unnecessary use of the blank identifier.
-        # https://staticcheck.dev/docs/checks/#S1005
-        - S1005
-        # Use "for { ... }" for infinite loops.
-        # https://staticcheck.dev/docs/checks/#S1006
-        - S1006
-        # Simplify regular expression by using raw string literal.
-        # https://staticcheck.dev/docs/checks/#S1007
-        - S1007
-        # Simplify returning boolean expression.
-        # https://staticcheck.dev/docs/checks/#S1008
-        - S1008
-        # Omit redundant nil check on slices, maps, and channels.
-        # https://staticcheck.dev/docs/checks/#S1009
-        - S1009
-        # Omit default slice index.
-        # https://staticcheck.dev/docs/checks/#S1010
-        - S1010
-        # Use a single 'append' to concatenate two slices.
-        # https://staticcheck.dev/docs/checks/#S1011
-        - S1011
-        # Replace 'time.Now().Sub(x)' with 'time.Since(x)'.
-        # https://staticcheck.dev/docs/checks/#S1012
-        - S1012
-        # Use a type conversion instead of manually copying struct fields.
-        # https://staticcheck.dev/docs/checks/#S1016
-        - S1016
-        # Replace manual trimming with 'strings.TrimPrefix'.
-        # https://staticcheck.dev/docs/checks/#S1017
-        - S1017
-        # Use "copy" for sliding elements.
-        # https://staticcheck.dev/docs/checks/#S1018
-        - S1018
-        # Simplify "make" call by omitting redundant arguments.
-        # https://staticcheck.dev/docs/checks/#S1019
-        - S1019
-        # Omit redundant nil check in type assertion.
-        # https://staticcheck.dev/docs/checks/#S1020
-        - S1020
-        # Merge variable declaration and assignment.
-        # https://staticcheck.dev/docs/checks/#S1021
-        - S1021
-        # Omit redundant control flow.
-        # https://staticcheck.dev/docs/checks/#S1023
-        - S1023
-        # Replace 'x.Sub(time.Now())' with 'time.Until(x)'.
-        # https://staticcheck.dev/docs/checks/#S1024
-        - S1024
-        # Don't use 'fmt.Sprintf("%s", x)' unnecessarily.
-        # https://staticcheck.dev/docs/checks/#S1025
-        - S1025
-        # Simplify error construction with 'fmt.Errorf'.
-        # https://staticcheck.dev/docs/checks/#S1028
-        - S1028
-        # Range over the string directly.
-        # https://staticcheck.dev/docs/checks/#S1029
-        - S1029
-        # Use 'bytes.Buffer.String' or 'bytes.Buffer.Bytes'.
-        # https://staticcheck.dev/docs/checks/#S1030
-        - S1030
-        # Omit redundant nil check around loop.
-        # https://staticcheck.dev/docs/checks/#S1031
-        - S1031
-        # Use 'sort.Ints(x)', 'sort.Float64s(x)', and 'sort.Strings(x)'.
-        # https://staticcheck.dev/docs/checks/#S1032
-        - S1032
-        # Unnecessary guard around call to "delete".
-        # https://staticcheck.dev/docs/checks/#S1033
-        - S1033
-        # Use result of type assertion to simplify cases.
-        # https://staticcheck.dev/docs/checks/#S1034
-        - S1034
-        # Redundant call to 'net/http.CanonicalHeaderKey' in method call on 'net/http.Header'.
-        # https://staticcheck.dev/docs/checks/#S1035
-        - S1035
-        # Unnecessary guard around map access.
-        # https://staticcheck.dev/docs/checks/#S1036
-        - S1036
-        # Elaborate way of sleeping.
-        # https://staticcheck.dev/docs/checks/#S1037
-        - S1037
-        # Unnecessarily complex way of printing formatted string.
-        # https://staticcheck.dev/docs/checks/#S1038
-        - S1038
-        # Unnecessary use of 'fmt.Sprint'.
-        # https://staticcheck.dev/docs/checks/#S1039
-        - S1039
-        # Type assertion to current type.
-        # https://staticcheck.dev/docs/checks/#S1040
-        - S1040
-        # Apply De Morgan's law.
-        # https://staticcheck.dev/docs/checks/#QF1001
-        - QF1001
-        # Convert untagged switch to tagged switch.
-        # https://staticcheck.dev/docs/checks/#QF1002
-        - QF1002
-        # Convert if/else-if chain to tagged switch.
-        # https://staticcheck.dev/docs/checks/#QF1003
-        - QF1003
-        # Use 'strings.ReplaceAll' instead of 'strings.Replace' with 'n == -1'.
-        # https://staticcheck.dev/docs/checks/#QF1004
-        - QF1004
-        # Expand call to 'math.Pow'.
-        # https://staticcheck.dev/docs/checks/#QF1005
-        - QF1005
-        # Lift 'if'+'break' into loop condition.
-        # https://staticcheck.dev/docs/checks/#QF1006
-        - QF1006
-        # Merge conditional assignment into variable declaration.
-        # https://staticcheck.dev/docs/checks/#QF1007
-        - QF1007
-        # Omit embedded fields from selector expression.
-        # https://staticcheck.dev/docs/checks/#QF1008
-        - QF1008
-        # Use 'time.Time.Equal' instead of '==' operator.
-        # https://staticcheck.dev/docs/checks/#QF1009
-        - QF1009
-        # Convert slice of bytes to string when printing it.
-        # https://staticcheck.dev/docs/checks/#QF1010
-        - QF1010
-        # Omit redundant type from variable declaration.
-        # https://staticcheck.dev/docs/checks/#QF1011
-        - QF1011
-        # Use 'fmt.Fprintf(x, ...)' instead of 'x.Write(fmt.Sprintf(...))'.
-        # https://staticcheck.dev/docs/checks/#QF1012
-        - QF1012
     unused:
-      # Mark all struct fields that have been written to as used.
       # Default: true
-      field-writes-are-uses: false
-      # Treat IncDec statement (e.g. `i++` or `i--`) as both read and write operation instead of just write.
+      field-writes-are-uses: true
       # Default: false
       post-statements-are-reads: true
-      # Mark all exported fields as used.
-      # default: true
-      exported-fields-are-used: false
-      # Mark all function parameters as used.
-      # default: true
-      parameters-are-used: true
-      # Mark all local variables as used.
-      # default: true
-      local-variables-are-used: false
-      # Mark all identifiers inside generated files as used.
       # Default: true
-      generated-is-used: false
+      exported-fields-are-used: true
+      # Default: true
+      parameters-are-used: true
+      # Default: true
+      local-variables-are-used: false
+      # Default: true — must be true, ent generates 130K+ lines of code
+      generated-is-used: true
 
 formatters:
   enable:

backend/cmd/server/wire_gen.go

@@ -164,7 +164,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	digestSessionStore := service.NewDigestSessionStore()
 	gatewayService := service.NewGatewayService(accountRepository, groupRepository, usageLogRepository, userRepository, userSubscriptionRepository, userGroupRateRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, identityService, httpUpstream, deferredService, claudeTokenProvider, sessionLimitCache, rpmCache, digestSessionStore)
 	openAITokenProvider := service.NewOpenAITokenProvider(accountRepository, geminiTokenCache, openAIOAuthService)
-	openAIGatewayService := service.NewOpenAIGatewayService(accountRepository, usageLogRepository, userRepository, userSubscriptionRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, httpUpstream, deferredService, openAITokenProvider)
+	openAIGatewayService := service.NewOpenAIGatewayService(accountRepository, usageLogRepository, userRepository, userSubscriptionRepository, userGroupRateRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, httpUpstream, deferredService, openAITokenProvider)
 	geminiMessagesCompatService := service.NewGeminiMessagesCompatService(accountRepository, groupRepository, gatewayCache, schedulerSnapshotService, geminiTokenProvider, rateLimitService, httpUpstream, antigravityGatewayService, configConfig)
 	opsSystemLogSink := service.ProvideOpsSystemLogSink(opsRepository)
 	opsService := service.NewOpsService(opsRepository, settingRepository, configConfig, accountRepository, userRepository, concurrencyService, gatewayService, openAIGatewayService, geminiMessagesCompatService, antigravityGatewayService, opsSystemLogSink)

backend/ent/announcement.go

@@ -25,6 +25,8 @@ type Announcement struct {
 	Content string `json:"content,omitempty"`
 	// 状态: draft, active, archived
 	Status string `json:"status,omitempty"`
+	// 通知模式: silent(仅铃铛), popup(弹窗提醒)
+	NotifyMode string `json:"notify_mode,omitempty"`
 	// 展示条件（JSON 规则）
 	Targeting domain.AnnouncementTargeting `json:"targeting,omitempty"`
 	// 开始展示时间（为空表示立即生效）
@@ -72,7 +74,7 @@ func (*Announcement) scanValues(columns []string) ([]any, error) {
 			values[i] = new([]byte)
 		case announcement.FieldID, announcement.FieldCreatedBy, announcement.FieldUpdatedBy:
 			values[i] = new(sql.NullInt64)
-		case announcement.FieldTitle, announcement.FieldContent, announcement.FieldStatus:
+		case announcement.FieldTitle, announcement.FieldContent, announcement.FieldStatus, announcement.FieldNotifyMode:
 			values[i] = new(sql.NullString)
 		case announcement.FieldStartsAt, announcement.FieldEndsAt, announcement.FieldCreatedAt, announcement.FieldUpdatedAt:
 			values[i] = new(sql.NullTime)
@@ -115,6 +117,12 @@ func (_m *Announcement) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.Status = value.String
 			}
+		case announcement.FieldNotifyMode:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field notify_mode", values[i])
+			} else if value.Valid {
+				_m.NotifyMode = value.String
+			}
 		case announcement.FieldTargeting:
 			if value, ok := values[i].(*[]byte); !ok {
 				return fmt.Errorf("unexpected type %T for field targeting", values[i])
@@ -213,6 +221,9 @@ func (_m *Announcement) String() string {
 	builder.WriteString("status=")
 	builder.WriteString(_m.Status)
 	builder.WriteString(", ")
+	builder.WriteString("notify_mode=")
+	builder.WriteString(_m.NotifyMode)
+	builder.WriteString(", ")
 	builder.WriteString("targeting=")
 	builder.WriteString(fmt.Sprintf("%v", _m.Targeting))
 	builder.WriteString(", ")

backend/ent/announcement/announcement.go

@@ -20,6 +20,8 @@ const (
 	FieldContent = "content"
 	// FieldStatus holds the string denoting the status field in the database.
 	FieldStatus = "status"
+	// FieldNotifyMode holds the string denoting the notify_mode field in the database.
+	FieldNotifyMode = "notify_mode"
 	// FieldTargeting holds the string denoting the targeting field in the database.
 	FieldTargeting = "targeting"
 	// FieldStartsAt holds the string denoting the starts_at field in the database.
@@ -53,6 +55,7 @@ var Columns = []string{
 	FieldTitle,
 	FieldContent,
 	FieldStatus,
+	FieldNotifyMode,
 	FieldTargeting,
 	FieldStartsAt,
 	FieldEndsAt,
@@ -81,6 +84,10 @@ var (
 	DefaultStatus string
 	// StatusValidator is a validator for the "status" field. It is called by the builders before save.
 	StatusValidator func(string) error
+	// DefaultNotifyMode holds the default value on creation for the "notify_mode" field.
+	DefaultNotifyMode string
+	// NotifyModeValidator is a validator for the "notify_mode" field. It is called by the builders before save.
+	NotifyModeValidator func(string) error
 	// DefaultCreatedAt holds the default value on creation for the "created_at" field.
 	DefaultCreatedAt func() time.Time
 	// DefaultUpdatedAt holds the default value on creation for the "updated_at" field.
@@ -112,6 +119,11 @@ func ByStatus(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldStatus, opts...).ToFunc()
 }
 
+// ByNotifyMode orders the results by the notify_mode field.
+func ByNotifyMode(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldNotifyMode, opts...).ToFunc()
+}
+
 // ByStartsAt orders the results by the starts_at field.
 func ByStartsAt(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldStartsAt, opts...).ToFunc()

backend/ent/announcement/where.go

@@ -70,6 +70,11 @@ func Status(v string) predicate.Announcement {
 	return predicate.Announcement(sql.FieldEQ(FieldStatus, v))
 }
 
+// NotifyMode applies equality check predicate on the "notify_mode" field. It's identical to NotifyModeEQ.
+func NotifyMode(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEQ(FieldNotifyMode, v))
+}
+
 // StartsAt applies equality check predicate on the "starts_at" field. It's identical to StartsAtEQ.
 func StartsAt(v time.Time) predicate.Announcement {
 	return predicate.Announcement(sql.FieldEQ(FieldStartsAt, v))
@@ -295,6 +300,71 @@ func StatusContainsFold(v string) predicate.Announcement {
 	return predicate.Announcement(sql.FieldContainsFold(FieldStatus, v))
 }
 
+// NotifyModeEQ applies the EQ predicate on the "notify_mode" field.
+func NotifyModeEQ(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEQ(FieldNotifyMode, v))
+}
+
+// NotifyModeNEQ applies the NEQ predicate on the "notify_mode" field.
+func NotifyModeNEQ(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldNEQ(FieldNotifyMode, v))
+}
+
+// NotifyModeIn applies the In predicate on the "notify_mode" field.
+func NotifyModeIn(vs ...string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldIn(FieldNotifyMode, vs...))
+}
+
+// NotifyModeNotIn applies the NotIn predicate on the "notify_mode" field.
+func NotifyModeNotIn(vs ...string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldNotIn(FieldNotifyMode, vs...))
+}
+
+// NotifyModeGT applies the GT predicate on the "notify_mode" field.
+func NotifyModeGT(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldGT(FieldNotifyMode, v))
+}
+
+// NotifyModeGTE applies the GTE predicate on the "notify_mode" field.
+func NotifyModeGTE(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldGTE(FieldNotifyMode, v))
+}
+
+// NotifyModeLT applies the LT predicate on the "notify_mode" field.
+func NotifyModeLT(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldLT(FieldNotifyMode, v))
+}
+
+// NotifyModeLTE applies the LTE predicate on the "notify_mode" field.
+func NotifyModeLTE(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldLTE(FieldNotifyMode, v))
+}
+
+// NotifyModeContains applies the Contains predicate on the "notify_mode" field.
+func NotifyModeContains(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldContains(FieldNotifyMode, v))
+}
+
+// NotifyModeHasPrefix applies the HasPrefix predicate on the "notify_mode" field.
+func NotifyModeHasPrefix(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldHasPrefix(FieldNotifyMode, v))
+}
+
+// NotifyModeHasSuffix applies the HasSuffix predicate on the "notify_mode" field.
+func NotifyModeHasSuffix(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldHasSuffix(FieldNotifyMode, v))
+}
+
+// NotifyModeEqualFold applies the EqualFold predicate on the "notify_mode" field.
+func NotifyModeEqualFold(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEqualFold(FieldNotifyMode, v))
+}
+
+// NotifyModeContainsFold applies the ContainsFold predicate on the "notify_mode" field.
+func NotifyModeContainsFold(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldContainsFold(FieldNotifyMode, v))
+}
+
 // TargetingIsNil applies the IsNil predicate on the "targeting" field.
 func TargetingIsNil() predicate.Announcement {
 	return predicate.Announcement(sql.FieldIsNull(FieldTargeting))

backend/ent/announcement_create.go

@@ -50,6 +50,20 @@ func (_c *AnnouncementCreate) SetNillableStatus(v *string) *AnnouncementCreate {
 	return _c
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_c *AnnouncementCreate) SetNotifyMode(v string) *AnnouncementCreate {
+	_c.mutation.SetNotifyMode(v)
+	return _c
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_c *AnnouncementCreate) SetNillableNotifyMode(v *string) *AnnouncementCreate {
+	if v != nil {
+		_c.SetNotifyMode(*v)
+	}
+	return _c
+}
+
 // SetTargeting sets the "targeting" field.
 func (_c *AnnouncementCreate) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementCreate {
 	_c.mutation.SetTargeting(v)
@@ -202,6 +216,10 @@ func (_c *AnnouncementCreate) defaults() {
 		v := announcement.DefaultStatus
 		_c.mutation.SetStatus(v)
 	}
+	if _, ok := _c.mutation.NotifyMode(); !ok {
+		v := announcement.DefaultNotifyMode
+		_c.mutation.SetNotifyMode(v)
+	}
 	if _, ok := _c.mutation.CreatedAt(); !ok {
 		v := announcement.DefaultCreatedAt()
 		_c.mutation.SetCreatedAt(v)
@@ -238,6 +256,14 @@ func (_c *AnnouncementCreate) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if _, ok := _c.mutation.NotifyMode(); !ok {
+		return &ValidationError{Name: "notify_mode", err: errors.New(`ent: missing required field "Announcement.notify_mode"`)}
+	}
+	if v, ok := _c.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	if _, ok := _c.mutation.CreatedAt(); !ok {
 		return &ValidationError{Name: "created_at", err: errors.New(`ent: missing required field "Announcement.created_at"`)}
 	}
@@ -283,6 +309,10 @@ func (_c *AnnouncementCreate) createSpec() (*Announcement, *sqlgraph.CreateSpec)
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 		_node.Status = value
 	}
+	if value, ok := _c.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+		_node.NotifyMode = value
+	}
 	if value, ok := _c.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 		_node.Targeting = value
@@ -415,6 +445,18 @@ func (u *AnnouncementUpsert) UpdateStatus() *AnnouncementUpsert {
 	return u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsert) SetNotifyMode(v string) *AnnouncementUpsert {
+	u.Set(announcement.FieldNotifyMode, v)
+	return u
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsert) UpdateNotifyMode() *AnnouncementUpsert {
+	u.SetExcluded(announcement.FieldNotifyMode)
+	return u
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsert) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsert {
 	u.Set(announcement.FieldTargeting, v)
@@ -616,6 +658,20 @@ func (u *AnnouncementUpsertOne) UpdateStatus() *AnnouncementUpsertOne {
 	})
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsertOne) SetNotifyMode(v string) *AnnouncementUpsertOne {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.SetNotifyMode(v)
+	})
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsertOne) UpdateNotifyMode() *AnnouncementUpsertOne {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.UpdateNotifyMode()
+	})
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsertOne) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsertOne {
 	return u.Update(func(s *AnnouncementUpsert) {
@@ -1002,6 +1058,20 @@ func (u *AnnouncementUpsertBulk) UpdateStatus() *AnnouncementUpsertBulk {
 	})
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsertBulk) SetNotifyMode(v string) *AnnouncementUpsertBulk {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.SetNotifyMode(v)
+	})
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsertBulk) UpdateNotifyMode() *AnnouncementUpsertBulk {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.UpdateNotifyMode()
+	})
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsertBulk) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsertBulk {
 	return u.Update(func(s *AnnouncementUpsert) {

backend/ent/announcement_update.go

@@ -72,6 +72,20 @@ func (_u *AnnouncementUpdate) SetNillableStatus(v *string) *AnnouncementUpdate {
 	return _u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_u *AnnouncementUpdate) SetNotifyMode(v string) *AnnouncementUpdate {
+	_u.mutation.SetNotifyMode(v)
+	return _u
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_u *AnnouncementUpdate) SetNillableNotifyMode(v *string) *AnnouncementUpdate {
+	if v != nil {
+		_u.SetNotifyMode(*v)
+	}
+	return _u
+}
+
 // SetTargeting sets the "targeting" field.
 func (_u *AnnouncementUpdate) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpdate {
 	_u.mutation.SetTargeting(v)
@@ -286,6 +300,11 @@ func (_u *AnnouncementUpdate) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -310,6 +329,9 @@ func (_u *AnnouncementUpdate) sqlSave(ctx context.Context) (_node int, err error
 	if value, ok := _u.mutation.Status(); ok {
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+	}
 	if value, ok := _u.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 	}
@@ -456,6 +478,20 @@ func (_u *AnnouncementUpdateOne) SetNillableStatus(v *string) *AnnouncementUpdat
 	return _u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_u *AnnouncementUpdateOne) SetNotifyMode(v string) *AnnouncementUpdateOne {
+	_u.mutation.SetNotifyMode(v)
+	return _u
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_u *AnnouncementUpdateOne) SetNillableNotifyMode(v *string) *AnnouncementUpdateOne {
+	if v != nil {
+		_u.SetNotifyMode(*v)
+	}
+	return _u
+}
+
 // SetTargeting sets the "targeting" field.
 func (_u *AnnouncementUpdateOne) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpdateOne {
 	_u.mutation.SetTargeting(v)
@@ -683,6 +719,11 @@ func (_u *AnnouncementUpdateOne) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -724,6 +765,9 @@ func (_u *AnnouncementUpdateOne) sqlSave(ctx context.Context) (_node *Announceme
 	if value, ok := _u.mutation.Status(); ok {
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+	}
 	if value, ok := _u.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 	}

backend/ent/group.go

@@ -78,6 +78,10 @@ type Group struct {
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`
 	// 分组显示排序，数值越小越靠前
 	SortOrder int `json:"sort_order,omitempty"`
+	// 是否允许 /v1/messages 调度到此 OpenAI 分组
+	AllowMessagesDispatch bool `json:"allow_messages_dispatch,omitempty"`
+	// 默认映射模型 ID，当账号级映射找不到时使用此值
+	DefaultMappedModel string `json:"default_mapped_model,omitempty"`
 	// Edges holds the relations/edges for other nodes in the graph.
 	// The values are being populated by the GroupQuery when eager-loading is set.
 	Edges        GroupEdges `json:"edges"`
@@ -186,13 +190,13 @@ func (*Group) scanValues(columns []string) ([]any, error) {
 		switch columns[i] {
 		case group.FieldModelRouting, group.FieldSupportedModelScopes:
 			values[i] = new([]byte)
-		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject:
+		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject, group.FieldAllowMessagesDispatch:
 			values[i] = new(sql.NullBool)
 		case group.FieldRateMultiplier, group.FieldDailyLimitUsd, group.FieldWeeklyLimitUsd, group.FieldMonthlyLimitUsd, group.FieldImagePrice1k, group.FieldImagePrice2k, group.FieldImagePrice4k, group.FieldSoraImagePrice360, group.FieldSoraImagePrice540, group.FieldSoraVideoPricePerRequest, group.FieldSoraVideoPricePerRequestHd:
 			values[i] = new(sql.NullFloat64)
 		case group.FieldID, group.FieldDefaultValidityDays, group.FieldSoraStorageQuotaBytes, group.FieldFallbackGroupID, group.FieldFallbackGroupIDOnInvalidRequest, group.FieldSortOrder:
 			values[i] = new(sql.NullInt64)
-		case group.FieldName, group.FieldDescription, group.FieldStatus, group.FieldPlatform, group.FieldSubscriptionType:
+		case group.FieldName, group.FieldDescription, group.FieldStatus, group.FieldPlatform, group.FieldSubscriptionType, group.FieldDefaultMappedModel:
 			values[i] = new(sql.NullString)
 		case group.FieldCreatedAt, group.FieldUpdatedAt, group.FieldDeletedAt:
 			values[i] = new(sql.NullTime)
@@ -415,6 +419,18 @@ func (_m *Group) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.SortOrder = int(value.Int64)
 			}
+		case group.FieldAllowMessagesDispatch:
+			if value, ok := values[i].(*sql.NullBool); !ok {
+				return fmt.Errorf("unexpected type %T for field allow_messages_dispatch", values[i])
+			} else if value.Valid {
+				_m.AllowMessagesDispatch = value.Bool
+			}
+		case group.FieldDefaultMappedModel:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field default_mapped_model", values[i])
+			} else if value.Valid {
+				_m.DefaultMappedModel = value.String
+			}
 		default:
 			_m.selectValues.Set(columns[i], values[i])
 		}
@@ -608,6 +624,12 @@ func (_m *Group) String() string {
 	builder.WriteString(", ")
 	builder.WriteString("sort_order=")
 	builder.WriteString(fmt.Sprintf("%v", _m.SortOrder))
+	builder.WriteString(", ")
+	builder.WriteString("allow_messages_dispatch=")
+	builder.WriteString(fmt.Sprintf("%v", _m.AllowMessagesDispatch))
+	builder.WriteString(", ")
+	builder.WriteString("default_mapped_model=")
+	builder.WriteString(_m.DefaultMappedModel)
 	builder.WriteByte(')')
 	return builder.String()
 }

backend/ent/group/group.go

@@ -75,6 +75,10 @@ const (
 	FieldSupportedModelScopes = "supported_model_scopes"
 	// FieldSortOrder holds the string denoting the sort_order field in the database.
 	FieldSortOrder = "sort_order"
+	// FieldAllowMessagesDispatch holds the string denoting the allow_messages_dispatch field in the database.
+	FieldAllowMessagesDispatch = "allow_messages_dispatch"
+	// FieldDefaultMappedModel holds the string denoting the default_mapped_model field in the database.
+	FieldDefaultMappedModel = "default_mapped_model"
 	// EdgeAPIKeys holds the string denoting the api_keys edge name in mutations.
 	EdgeAPIKeys = "api_keys"
 	// EdgeRedeemCodes holds the string denoting the redeem_codes edge name in mutations.
@@ -180,6 +184,8 @@ var Columns = []string{
 	FieldMcpXMLInject,
 	FieldSupportedModelScopes,
 	FieldSortOrder,
+	FieldAllowMessagesDispatch,
+	FieldDefaultMappedModel,
 }
 
 var (
@@ -247,6 +253,12 @@ var (
 	DefaultSupportedModelScopes []string
 	// DefaultSortOrder holds the default value on creation for the "sort_order" field.
 	DefaultSortOrder int
+	// DefaultAllowMessagesDispatch holds the default value on creation for the "allow_messages_dispatch" field.
+	DefaultAllowMessagesDispatch bool
+	// DefaultDefaultMappedModel holds the default value on creation for the "default_mapped_model" field.
+	DefaultDefaultMappedModel string
+	// DefaultMappedModelValidator is a validator for the "default_mapped_model" field. It is called by the builders before save.
+	DefaultMappedModelValidator func(string) error
 )
 
 // OrderOption defines the ordering options for the Group queries.
@@ -397,6 +409,16 @@ func BySortOrder(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldSortOrder, opts...).ToFunc()
 }
 
+// ByAllowMessagesDispatch orders the results by the allow_messages_dispatch field.
+func ByAllowMessagesDispatch(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldAllowMessagesDispatch, opts...).ToFunc()
+}
+
+// ByDefaultMappedModel orders the results by the default_mapped_model field.
+func ByDefaultMappedModel(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldDefaultMappedModel, opts...).ToFunc()
+}
+
 // ByAPIKeysCount orders the results by api_keys count.
 func ByAPIKeysCount(opts ...sql.OrderTermOption) OrderOption {
 	return func(s *sql.Selector) {

backend/ent/group/where.go

@@ -195,6 +195,16 @@ func SortOrder(v int) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldSortOrder, v))
 }
 
+// AllowMessagesDispatch applies equality check predicate on the "allow_messages_dispatch" field. It's identical to AllowMessagesDispatchEQ.
+func AllowMessagesDispatch(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldAllowMessagesDispatch, v))
+}
+
+// DefaultMappedModel applies equality check predicate on the "default_mapped_model" field. It's identical to DefaultMappedModelEQ.
+func DefaultMappedModel(v string) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldDefaultMappedModel, v))
+}
+
 // CreatedAtEQ applies the EQ predicate on the "created_at" field.
 func CreatedAtEQ(v time.Time) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldCreatedAt, v))
@@ -1470,6 +1480,81 @@ func SortOrderLTE(v int) predicate.Group {
 	return predicate.Group(sql.FieldLTE(FieldSortOrder, v))
 }
 
+// AllowMessagesDispatchEQ applies the EQ predicate on the "allow_messages_dispatch" field.
+func AllowMessagesDispatchEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldAllowMessagesDispatch, v))
+}
+
+// AllowMessagesDispatchNEQ applies the NEQ predicate on the "allow_messages_dispatch" field.
+func AllowMessagesDispatchNEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldNEQ(FieldAllowMessagesDispatch, v))
+}
+
+// DefaultMappedModelEQ applies the EQ predicate on the "default_mapped_model" field.
+func DefaultMappedModelEQ(v string) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelNEQ applies the NEQ predicate on the "default_mapped_model" field.
+func DefaultMappedModelNEQ(v string) predicate.Group {
+	return predicate.Group(sql.FieldNEQ(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelIn applies the In predicate on the "default_mapped_model" field.
+func DefaultMappedModelIn(vs ...string) predicate.Group {
+	return predicate.Group(sql.FieldIn(FieldDefaultMappedModel, vs...))
+}
+
+// DefaultMappedModelNotIn applies the NotIn predicate on the "default_mapped_model" field.
+func DefaultMappedModelNotIn(vs ...string) predicate.Group {
+	return predicate.Group(sql.FieldNotIn(FieldDefaultMappedModel, vs...))
+}
+
+// DefaultMappedModelGT applies the GT predicate on the "default_mapped_model" field.
+func DefaultMappedModelGT(v string) predicate.Group {
+	return predicate.Group(sql.FieldGT(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelGTE applies the GTE predicate on the "default_mapped_model" field.
+func DefaultMappedModelGTE(v string) predicate.Group {
+	return predicate.Group(sql.FieldGTE(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelLT applies the LT predicate on the "default_mapped_model" field.
+func DefaultMappedModelLT(v string) predicate.Group {
+	return predicate.Group(sql.FieldLT(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelLTE applies the LTE predicate on the "default_mapped_model" field.
+func DefaultMappedModelLTE(v string) predicate.Group {
+	return predicate.Group(sql.FieldLTE(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelContains applies the Contains predicate on the "default_mapped_model" field.
+func DefaultMappedModelContains(v string) predicate.Group {
+	return predicate.Group(sql.FieldContains(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelHasPrefix applies the HasPrefix predicate on the "default_mapped_model" field.
+func DefaultMappedModelHasPrefix(v string) predicate.Group {
+	return predicate.Group(sql.FieldHasPrefix(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelHasSuffix applies the HasSuffix predicate on the "default_mapped_model" field.
+func DefaultMappedModelHasSuffix(v string) predicate.Group {
+	return predicate.Group(sql.FieldHasSuffix(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelEqualFold applies the EqualFold predicate on the "default_mapped_model" field.
+func DefaultMappedModelEqualFold(v string) predicate.Group {
+	return predicate.Group(sql.FieldEqualFold(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelContainsFold applies the ContainsFold predicate on the "default_mapped_model" field.
+func DefaultMappedModelContainsFold(v string) predicate.Group {
+	return predicate.Group(sql.FieldContainsFold(FieldDefaultMappedModel, v))
+}
+
 // HasAPIKeys applies the HasEdge predicate on the "api_keys" edge.
 func HasAPIKeys() predicate.Group {
 	return predicate.Group(func(s *sql.Selector) {

backend/ent/group_create.go

@@ -424,6 +424,34 @@ func (_c *GroupCreate) SetNillableSortOrder(v *int) *GroupCreate {
 	return _c
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_c *GroupCreate) SetAllowMessagesDispatch(v bool) *GroupCreate {
+	_c.mutation.SetAllowMessagesDispatch(v)
+	return _c
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_c *GroupCreate) SetNillableAllowMessagesDispatch(v *bool) *GroupCreate {
+	if v != nil {
+		_c.SetAllowMessagesDispatch(*v)
+	}
+	return _c
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_c *GroupCreate) SetDefaultMappedModel(v string) *GroupCreate {
+	_c.mutation.SetDefaultMappedModel(v)
+	return _c
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_c *GroupCreate) SetNillableDefaultMappedModel(v *string) *GroupCreate {
+	if v != nil {
+		_c.SetDefaultMappedModel(*v)
+	}
+	return _c
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_c *GroupCreate) AddAPIKeyIDs(ids ...int64) *GroupCreate {
 	_c.mutation.AddAPIKeyIDs(ids...)
@@ -613,6 +641,14 @@ func (_c *GroupCreate) defaults() error {
 		v := group.DefaultSortOrder
 		_c.mutation.SetSortOrder(v)
 	}
+	if _, ok := _c.mutation.AllowMessagesDispatch(); !ok {
+		v := group.DefaultAllowMessagesDispatch
+		_c.mutation.SetAllowMessagesDispatch(v)
+	}
+	if _, ok := _c.mutation.DefaultMappedModel(); !ok {
+		v := group.DefaultDefaultMappedModel
+		_c.mutation.SetDefaultMappedModel(v)
+	}
 	return nil
 }
 
@@ -683,6 +719,17 @@ func (_c *GroupCreate) check() error {
 	if _, ok := _c.mutation.SortOrder(); !ok {
 		return &ValidationError{Name: "sort_order", err: errors.New(`ent: missing required field "Group.sort_order"`)}
 	}
+	if _, ok := _c.mutation.AllowMessagesDispatch(); !ok {
+		return &ValidationError{Name: "allow_messages_dispatch", err: errors.New(`ent: missing required field "Group.allow_messages_dispatch"`)}
+	}
+	if _, ok := _c.mutation.DefaultMappedModel(); !ok {
+		return &ValidationError{Name: "default_mapped_model", err: errors.New(`ent: missing required field "Group.default_mapped_model"`)}
+	}
+	if v, ok := _c.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -830,6 +877,14 @@ func (_c *GroupCreate) createSpec() (*Group, *sqlgraph.CreateSpec) {
 		_spec.SetField(group.FieldSortOrder, field.TypeInt, value)
 		_node.SortOrder = value
 	}
+	if value, ok := _c.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+		_node.AllowMessagesDispatch = value
+	}
+	if value, ok := _c.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+		_node.DefaultMappedModel = value
+	}
 	if nodes := _c.mutation.APIKeysIDs(); len(nodes) > 0 {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -1520,6 +1575,30 @@ func (u *GroupUpsert) AddSortOrder(v int) *GroupUpsert {
 	return u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsert) SetAllowMessagesDispatch(v bool) *GroupUpsert {
+	u.Set(group.FieldAllowMessagesDispatch, v)
+	return u
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsert) UpdateAllowMessagesDispatch() *GroupUpsert {
+	u.SetExcluded(group.FieldAllowMessagesDispatch)
+	return u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsert) SetDefaultMappedModel(v string) *GroupUpsert {
+	u.Set(group.FieldDefaultMappedModel, v)
+	return u
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsert) UpdateDefaultMappedModel() *GroupUpsert {
+	u.SetExcluded(group.FieldDefaultMappedModel)
+	return u
+}
+
 // UpdateNewValues updates the mutable fields using the new values that were set on create.
 // Using this option is equivalent to using:
 //
@@ -2188,6 +2267,34 @@ func (u *GroupUpsertOne) UpdateSortOrder() *GroupUpsertOne {
 	})
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsertOne) SetAllowMessagesDispatch(v bool) *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetAllowMessagesDispatch(v)
+	})
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsertOne) UpdateAllowMessagesDispatch() *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateAllowMessagesDispatch()
+	})
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsertOne) SetDefaultMappedModel(v string) *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetDefaultMappedModel(v)
+	})
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsertOne) UpdateDefaultMappedModel() *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateDefaultMappedModel()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertOne) Exec(ctx context.Context) error {
 	if len(u.create.conflict) == 0 {
@@ -3022,6 +3129,34 @@ func (u *GroupUpsertBulk) UpdateSortOrder() *GroupUpsertBulk {
 	})
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsertBulk) SetAllowMessagesDispatch(v bool) *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetAllowMessagesDispatch(v)
+	})
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsertBulk) UpdateAllowMessagesDispatch() *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateAllowMessagesDispatch()
+	})
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsertBulk) SetDefaultMappedModel(v string) *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetDefaultMappedModel(v)
+	})
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsertBulk) UpdateDefaultMappedModel() *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateDefaultMappedModel()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertBulk) Exec(ctx context.Context) error {
 	if u.create.err != nil {

backend/ent/group_update.go

@@ -625,6 +625,34 @@ func (_u *GroupUpdate) AddSortOrder(v int) *GroupUpdate {
 	return _u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_u *GroupUpdate) SetAllowMessagesDispatch(v bool) *GroupUpdate {
+	_u.mutation.SetAllowMessagesDispatch(v)
+	return _u
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_u *GroupUpdate) SetNillableAllowMessagesDispatch(v *bool) *GroupUpdate {
+	if v != nil {
+		_u.SetAllowMessagesDispatch(*v)
+	}
+	return _u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_u *GroupUpdate) SetDefaultMappedModel(v string) *GroupUpdate {
+	_u.mutation.SetDefaultMappedModel(v)
+	return _u
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_u *GroupUpdate) SetNillableDefaultMappedModel(v *string) *GroupUpdate {
+	if v != nil {
+		_u.SetDefaultMappedModel(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdate) AddAPIKeyIDs(ids ...int64) *GroupUpdate {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -910,6 +938,11 @@ func (_u *GroupUpdate) check() error {
 			return &ValidationError{Name: "subscription_type", err: fmt.Errorf(`ent: validator failed for field "Group.subscription_type": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -1110,6 +1143,12 @@ func (_u *GroupUpdate) sqlSave(ctx context.Context) (_node int, err error) {
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+	}
+	if value, ok := _u.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -2014,6 +2053,34 @@ func (_u *GroupUpdateOne) AddSortOrder(v int) *GroupUpdateOne {
 	return _u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_u *GroupUpdateOne) SetAllowMessagesDispatch(v bool) *GroupUpdateOne {
+	_u.mutation.SetAllowMessagesDispatch(v)
+	return _u
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_u *GroupUpdateOne) SetNillableAllowMessagesDispatch(v *bool) *GroupUpdateOne {
+	if v != nil {
+		_u.SetAllowMessagesDispatch(*v)
+	}
+	return _u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_u *GroupUpdateOne) SetDefaultMappedModel(v string) *GroupUpdateOne {
+	_u.mutation.SetDefaultMappedModel(v)
+	return _u
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_u *GroupUpdateOne) SetNillableDefaultMappedModel(v *string) *GroupUpdateOne {
+	if v != nil {
+		_u.SetDefaultMappedModel(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdateOne) AddAPIKeyIDs(ids ...int64) *GroupUpdateOne {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -2312,6 +2379,11 @@ func (_u *GroupUpdateOne) check() error {
 			return &ValidationError{Name: "subscription_type", err: fmt.Errorf(`ent: validator failed for field "Group.subscription_type": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -2529,6 +2601,12 @@ func (_u *GroupUpdateOne) sqlSave(ctx context.Context) (_node *Group, err error)
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+	}
+	if value, ok := _u.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,

backend/ent/migrate/schema.go

@@ -251,6 +251,7 @@ var (
 		{Name: "title", Type: field.TypeString, Size: 200},
 		{Name: "content", Type: field.TypeString, SchemaType: map[string]string{"postgres": "text"}},
 		{Name: "status", Type: field.TypeString, Size: 20, Default: "draft"},
+		{Name: "notify_mode", Type: field.TypeString, Size: 20, Default: "silent"},
 		{Name: "targeting", Type: field.TypeJSON, Nullable: true, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "starts_at", Type: field.TypeTime, Nullable: true, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "ends_at", Type: field.TypeTime, Nullable: true, SchemaType: map[string]string{"postgres": "timestamptz"}},
@@ -273,17 +274,17 @@ var (
 			{
 				Name:    "announcement_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[9]},
+				Columns: []*schema.Column{AnnouncementsColumns[10]},
 			},
 			{
 				Name:    "announcement_starts_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[5]},
+				Columns: []*schema.Column{AnnouncementsColumns[6]},
 			},
 			{
 				Name:    "announcement_ends_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[6]},
+				Columns: []*schema.Column{AnnouncementsColumns[7]},
 			},
 		},
 	}
@@ -407,6 +408,8 @@ var (
 		{Name: "mcp_xml_inject", Type: field.TypeBool, Default: true},
 		{Name: "supported_model_scopes", Type: field.TypeJSON, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "sort_order", Type: field.TypeInt, Default: 0},
+		{Name: "allow_messages_dispatch", Type: field.TypeBool, Default: false},
+		{Name: "default_mapped_model", Type: field.TypeString, Size: 100, Default: ""},
 	}
 	// GroupsTable holds the schema information for the "groups" table.
 	GroupsTable = &schema.Table{

backend/ent/mutation.go

@@ -5167,6 +5167,7 @@ type AnnouncementMutation struct {
 	title         *string
 	content       *string
 	status        *string
+	notify_mode   *string
 	targeting     *domain.AnnouncementTargeting
 	starts_at     *time.Time
 	ends_at       *time.Time
@@ -5391,6 +5392,42 @@ func (m *AnnouncementMutation) ResetStatus() {
 	m.status = nil
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (m *AnnouncementMutation) SetNotifyMode(s string) {
+	m.notify_mode = &s
+}
+
+// NotifyMode returns the value of the "notify_mode" field in the mutation.
+func (m *AnnouncementMutation) NotifyMode() (r string, exists bool) {
+	v := m.notify_mode
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldNotifyMode returns the old "notify_mode" field's value of the Announcement entity.
+// If the Announcement object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *AnnouncementMutation) OldNotifyMode(ctx context.Context) (v string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldNotifyMode is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldNotifyMode requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldNotifyMode: %w", err)
+	}
+	return oldValue.NotifyMode, nil
+}
+
+// ResetNotifyMode resets all changes to the "notify_mode" field.
+func (m *AnnouncementMutation) ResetNotifyMode() {
+	m.notify_mode = nil
+}
+
 // SetTargeting sets the "targeting" field.
 func (m *AnnouncementMutation) SetTargeting(dt domain.AnnouncementTargeting) {
 	m.targeting = &dt
@@ -5838,7 +5875,7 @@ func (m *AnnouncementMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *AnnouncementMutation) Fields() []string {
-	fields := make([]string, 0, 10)
+	fields := make([]string, 0, 11)
 	if m.title != nil {
 		fields = append(fields, announcement.FieldTitle)
 	}
@@ -5848,6 +5885,9 @@ func (m *AnnouncementMutation) Fields() []string {
 	if m.status != nil {
 		fields = append(fields, announcement.FieldStatus)
 	}
+	if m.notify_mode != nil {
+		fields = append(fields, announcement.FieldNotifyMode)
+	}
 	if m.targeting != nil {
 		fields = append(fields, announcement.FieldTargeting)
 	}
@@ -5883,6 +5923,8 @@ func (m *AnnouncementMutation) Field(name string) (ent.Value, bool) {
 		return m.Content()
 	case announcement.FieldStatus:
 		return m.Status()
+	case announcement.FieldNotifyMode:
+		return m.NotifyMode()
 	case announcement.FieldTargeting:
 		return m.Targeting()
 	case announcement.FieldStartsAt:
@@ -5912,6 +5954,8 @@ func (m *AnnouncementMutation) OldField(ctx context.Context, name string) (ent.V
 		return m.OldContent(ctx)
 	case announcement.FieldStatus:
 		return m.OldStatus(ctx)
+	case announcement.FieldNotifyMode:
+		return m.OldNotifyMode(ctx)
 	case announcement.FieldTargeting:
 		return m.OldTargeting(ctx)
 	case announcement.FieldStartsAt:
@@ -5956,6 +6000,13 @@ func (m *AnnouncementMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetStatus(v)
 		return nil
+	case announcement.FieldNotifyMode:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetNotifyMode(v)
+		return nil
 	case announcement.FieldTargeting:
 		v, ok := value.(domain.AnnouncementTargeting)
 		if !ok {
@@ -6123,6 +6174,9 @@ func (m *AnnouncementMutation) ResetField(name string) error {
 	case announcement.FieldStatus:
 		m.ResetStatus()
 		return nil
+	case announcement.FieldNotifyMode:
+		m.ResetNotifyMode()
+		return nil
 	case announcement.FieldTargeting:
 		m.ResetTargeting()
 		return nil
@@ -8196,6 +8250,8 @@ type GroupMutation struct {
 	appendsupported_model_scopes            []string
 	sort_order                              *int
 	addsort_order                           *int
+	allow_messages_dispatch                 *bool
+	default_mapped_model                    *string
 	clearedFields                           map[string]struct{}
 	api_keys                                map[int64]struct{}
 	removedapi_keys                         map[int64]struct{}
@@ -9940,6 +9996,78 @@ func (m *GroupMutation) ResetSortOrder() {
 	m.addsort_order = nil
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (m *GroupMutation) SetAllowMessagesDispatch(b bool) {
+	m.allow_messages_dispatch = &b
+}
+
+// AllowMessagesDispatch returns the value of the "allow_messages_dispatch" field in the mutation.
+func (m *GroupMutation) AllowMessagesDispatch() (r bool, exists bool) {
+	v := m.allow_messages_dispatch
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldAllowMessagesDispatch returns the old "allow_messages_dispatch" field's value of the Group entity.
+// If the Group object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *GroupMutation) OldAllowMessagesDispatch(ctx context.Context) (v bool, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldAllowMessagesDispatch is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldAllowMessagesDispatch requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldAllowMessagesDispatch: %w", err)
+	}
+	return oldValue.AllowMessagesDispatch, nil
+}
+
+// ResetAllowMessagesDispatch resets all changes to the "allow_messages_dispatch" field.
+func (m *GroupMutation) ResetAllowMessagesDispatch() {
+	m.allow_messages_dispatch = nil
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (m *GroupMutation) SetDefaultMappedModel(s string) {
+	m.default_mapped_model = &s
+}
+
+// DefaultMappedModel returns the value of the "default_mapped_model" field in the mutation.
+func (m *GroupMutation) DefaultMappedModel() (r string, exists bool) {
+	v := m.default_mapped_model
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldDefaultMappedModel returns the old "default_mapped_model" field's value of the Group entity.
+// If the Group object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *GroupMutation) OldDefaultMappedModel(ctx context.Context) (v string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldDefaultMappedModel is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldDefaultMappedModel requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldDefaultMappedModel: %w", err)
+	}
+	return oldValue.DefaultMappedModel, nil
+}
+
+// ResetDefaultMappedModel resets all changes to the "default_mapped_model" field.
+func (m *GroupMutation) ResetDefaultMappedModel() {
+	m.default_mapped_model = nil
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by ids.
 func (m *GroupMutation) AddAPIKeyIDs(ids ...int64) {
 	if m.api_keys == nil {
@@ -10298,7 +10426,7 @@ func (m *GroupMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *GroupMutation) Fields() []string {
-	fields := make([]string, 0, 31)
+	fields := make([]string, 0, 32)
 	if m.created_at != nil {
 		fields = append(fields, group.FieldCreatedAt)
 	}
@@ -10389,6 +10517,12 @@ func (m *GroupMutation) Fields() []string {
 	if m.sort_order != nil {
 		fields = append(fields, group.FieldSortOrder)
 	}
+	if m.allow_messages_dispatch != nil {
+		fields = append(fields, group.FieldAllowMessagesDispatch)
+	}
+	if m.default_mapped_model != nil {
+		fields = append(fields, group.FieldDefaultMappedModel)
+	}
 	return fields
 }
 
@@ -10457,6 +10591,10 @@ func (m *GroupMutation) Field(name string) (ent.Value, bool) {
 		return m.SupportedModelScopes()
 	case group.FieldSortOrder:
 		return m.SortOrder()
+	case group.FieldAllowMessagesDispatch:
+		return m.AllowMessagesDispatch()
+	case group.FieldDefaultMappedModel:
+		return m.DefaultMappedModel()
 	}
 	return nil, false
 }
@@ -10526,6 +10664,10 @@ func (m *GroupMutation) OldField(ctx context.Context, name string) (ent.Value, e
 		return m.OldSupportedModelScopes(ctx)
 	case group.FieldSortOrder:
 		return m.OldSortOrder(ctx)
+	case group.FieldAllowMessagesDispatch:
+		return m.OldAllowMessagesDispatch(ctx)
+	case group.FieldDefaultMappedModel:
+		return m.OldDefaultMappedModel(ctx)
 	}
 	return nil, fmt.Errorf("unknown Group field %s", name)
 }
@@ -10745,6 +10887,20 @@ func (m *GroupMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetSortOrder(v)
 		return nil
+	case group.FieldAllowMessagesDispatch:
+		v, ok := value.(bool)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetAllowMessagesDispatch(v)
+		return nil
+	case group.FieldDefaultMappedModel:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetDefaultMappedModel(v)
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }
@@ -11172,6 +11328,12 @@ func (m *GroupMutation) ResetField(name string) error {
 	case group.FieldSortOrder:
 		m.ResetSortOrder()
 		return nil
+	case group.FieldAllowMessagesDispatch:
+		m.ResetAllowMessagesDispatch()
+		return nil
+	case group.FieldDefaultMappedModel:
+		m.ResetDefaultMappedModel()
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }

backend/ent/runtime/runtime.go

@@ -277,12 +277,18 @@ func init() {
 	announcement.DefaultStatus = announcementDescStatus.Default.(string)
 	// announcement.StatusValidator is a validator for the "status" field. It is called by the builders before save.
 	announcement.StatusValidator = announcementDescStatus.Validators[0].(func(string) error)
+	// announcementDescNotifyMode is the schema descriptor for notify_mode field.
+	announcementDescNotifyMode := announcementFields[3].Descriptor()
+	// announcement.DefaultNotifyMode holds the default value on creation for the notify_mode field.
+	announcement.DefaultNotifyMode = announcementDescNotifyMode.Default.(string)
+	// announcement.NotifyModeValidator is a validator for the "notify_mode" field. It is called by the builders before save.
+	announcement.NotifyModeValidator = announcementDescNotifyMode.Validators[0].(func(string) error)
 	// announcementDescCreatedAt is the schema descriptor for created_at field.
-	announcementDescCreatedAt := announcementFields[8].Descriptor()
+	announcementDescCreatedAt := announcementFields[9].Descriptor()
 	// announcement.DefaultCreatedAt holds the default value on creation for the created_at field.
 	announcement.DefaultCreatedAt = announcementDescCreatedAt.Default.(func() time.Time)
 	// announcementDescUpdatedAt is the schema descriptor for updated_at field.
-	announcementDescUpdatedAt := announcementFields[9].Descriptor()
+	announcementDescUpdatedAt := announcementFields[10].Descriptor()
 	// announcement.DefaultUpdatedAt holds the default value on creation for the updated_at field.
 	announcement.DefaultUpdatedAt = announcementDescUpdatedAt.Default.(func() time.Time)
 	// announcement.UpdateDefaultUpdatedAt holds the default value on update for the updated_at field.
@@ -447,6 +453,16 @@ func init() {
 	groupDescSortOrder := groupFields[26].Descriptor()
 	// group.DefaultSortOrder holds the default value on creation for the sort_order field.
 	group.DefaultSortOrder = groupDescSortOrder.Default.(int)
+	// groupDescAllowMessagesDispatch is the schema descriptor for allow_messages_dispatch field.
+	groupDescAllowMessagesDispatch := groupFields[27].Descriptor()
+	// group.DefaultAllowMessagesDispatch holds the default value on creation for the allow_messages_dispatch field.
+	group.DefaultAllowMessagesDispatch = groupDescAllowMessagesDispatch.Default.(bool)
+	// groupDescDefaultMappedModel is the schema descriptor for default_mapped_model field.
+	groupDescDefaultMappedModel := groupFields[28].Descriptor()
+	// group.DefaultDefaultMappedModel holds the default value on creation for the default_mapped_model field.
+	group.DefaultDefaultMappedModel = groupDescDefaultMappedModel.Default.(string)
+	// group.DefaultMappedModelValidator is a validator for the "default_mapped_model" field. It is called by the builders before save.
+	group.DefaultMappedModelValidator = groupDescDefaultMappedModel.Validators[0].(func(string) error)
 	idempotencyrecordMixin := schema.IdempotencyRecord{}.Mixin()
 	idempotencyrecordMixinFields0 := idempotencyrecordMixin[0].Fields()
 	_ = idempotencyrecordMixinFields0

backend/ent/schema/announcement.go

@@ -41,6 +41,10 @@ func (Announcement) Fields() []ent.Field {
 			MaxLen(20).
 			Default(domain.AnnouncementStatusDraft).
 			Comment("状态: draft, active, archived"),
+		field.String("notify_mode").
+			MaxLen(20).
+			Default(domain.AnnouncementNotifyModeSilent).
+			Comment("通知模式: silent(仅铃铛), popup(弹窗提醒)"),
 		field.JSON("targeting", domain.AnnouncementTargeting{}).
 			Optional().
 			SchemaType(map[string]string{dialect.Postgres: "jsonb"}).

backend/ent/schema/group.go

@@ -148,6 +148,15 @@ func (Group) Fields() []ent.Field {
 		field.Int("sort_order").
 			Default(0).
 			Comment("分组显示排序，数值越小越靠前"),
+
+		// OpenAI Messages 调度配置 (added by migration 069)
+		field.Bool("allow_messages_dispatch").
+			Default(false).
+			Comment("是否允许 /v1/messages 调度到此 OpenAI 分组"),
+		field.String("default_mapped_model").
+			MaxLen(100).
+			Default("").
+			Comment("默认映射模型 ID，当账号级映射找不到时使用此值"),
 	}
 }
 

backend/go.mod

@@ -1,12 +1,13 @@
 module github.com/Wei-Shaw/sub2api
 
-go 1.25.7
+go 1.26.1
 
 require (
 	entgo.io/ent v0.14.5
 	github.com/DATA-DOG/go-sqlmock v1.5.2
 	github.com/DouDOU-start/go-sora2api v1.1.0
 	github.com/alitto/pond/v2 v2.6.2
+	github.com/aws/aws-sdk-go-v2 v1.41.2
 	github.com/aws/aws-sdk-go-v2/config v1.32.10
 	github.com/aws/aws-sdk-go-v2/credentials v1.19.10
 	github.com/aws/aws-sdk-go-v2/service/s3 v1.96.2
@@ -38,8 +39,6 @@ require (
 	golang.org/x/net v0.49.0
 	golang.org/x/sync v0.19.0
 	golang.org/x/term v0.40.0
-	google.golang.org/grpc v1.75.1
-	google.golang.org/protobuf v1.36.10
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v3 v3.0.1
 	modernc.org/sqlite v1.44.3
@@ -53,7 +52,6 @@ require (
 	github.com/agext/levenshtein v1.2.3 // indirect
 	github.com/andybalholm/brotli v1.2.0 // indirect
 	github.com/apparentlymart/go-textseg/v15 v15.0.0 // indirect
-	github.com/aws/aws-sdk-go-v2 v1.41.2 // indirect
 	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.5 // indirect
 	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.18 // indirect
 	github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.18 // indirect
@@ -109,7 +107,6 @@ require (
 	github.com/goccy/go-json v0.10.2 // indirect
 	github.com/google/go-cmp v0.7.0 // indirect
 	github.com/google/go-querystring v1.1.0 // indirect
-	github.com/google/subcommands v1.2.0 // indirect
 	github.com/grpc-ecosystem/grpc-gateway/v2 v2.27.3 // indirect
 	github.com/hashicorp/hcl v1.0.0 // indirect
 	github.com/hashicorp/hcl/v2 v2.18.1 // indirect
@@ -169,6 +166,7 @@ require (
 	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.49.0 // indirect
 	go.opentelemetry.io/otel v1.37.0 // indirect
 	go.opentelemetry.io/otel/metric v1.37.0 // indirect
+	go.opentelemetry.io/otel/sdk v1.37.0 // indirect
 	go.opentelemetry.io/otel/trace v1.37.0 // indirect
 	go.uber.org/atomic v1.10.0 // indirect
 	go.uber.org/automaxprocs v1.6.0 // indirect
@@ -178,8 +176,8 @@ require (
 	golang.org/x/mod v0.32.0 // indirect
 	golang.org/x/sys v0.41.0 // indirect
 	golang.org/x/text v0.34.0 // indirect
-	golang.org/x/tools v0.41.0 // indirect
-	google.golang.org/genproto/googleapis/rpc v0.0.0-20250929231259-57b25ae835d4 // indirect
+	google.golang.org/grpc v1.75.1 // indirect
+	google.golang.org/protobuf v1.36.10 // indirect
 	gopkg.in/ini.v1 v1.67.0 // indirect
 	modernc.org/libc v1.67.6 // indirect
 	modernc.org/mathutil v1.7.1 // indirect

backend/go.sum

@@ -124,8 +124,6 @@ github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f h1:lO4WD4F/r
 github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f/go.mod h1:cuUVRXasLTGF7a8hSLbxyZXjz+1KgoB3wDUb6vlszIc=
 github.com/distribution/reference v0.6.0 h1:0IXCQ5g4/QMHHkarYzh5l+u8T3t73zM5QvfrDyIgxBk=
 github.com/distribution/reference v0.6.0/go.mod h1:BbU0aIcezP1/5jX/8MP0YiH4SdvB5Y4f/wlDRiLyi3E=
-github.com/dlclark/regexp2 v1.10.0 h1:+/GIL799phkJqYW+3YbOd8LCcbHzT0Pbo8zl70MHsq0=
-github.com/dlclark/regexp2 v1.10.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
 github.com/docker/docker v28.5.1+incompatible h1:Bm8DchhSD2J6PsFzxC35TZo4TLGR2PdW/E69rU45NhM=
 github.com/docker/docker v28.5.1+incompatible/go.mod h1:eEKB0N0r5NX/I1kEveEz05bcu8tLC/8azJZsviup8Sk=
 github.com/docker/go-connections v0.6.0 h1:LlMG9azAe1TqfR7sO+NJttz1gy6KO7VJBh+pMmjSD94=
@@ -182,7 +180,6 @@ github.com/google/go-querystring v1.1.0/go.mod h1:Kcdr2DB4koayq7X8pmAG4sNG59So17
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e h1:ijClszYn+mADRFY17kjQEVQ1XRhq2/JR1M3sGqeJoxs=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e/go.mod h1:boTsfXsheKC2y+lKOCMpSfarhxDeIzfZG1jqGcPl3cA=
-github.com/google/subcommands v1.2.0/go.mod h1:ZjhPrFU+Olkh9WazFPsl27BQ4UPiG37m3yTrtFlrHVk=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/google/wire v0.7.0 h1:JxUKI6+CVBgCO2WToKy/nQk0sS+amI9z9EjVmdaocj4=
@@ -202,8 +199,6 @@ github.com/icholy/digest v1.1.0 h1:HfGg9Irj7i+IX1o1QAmPfIBNu/Q5A5Tu3n/MED9k9H4=
 github.com/icholy/digest v1.1.0/go.mod h1:QNrsSGQ5v7v9cReDI0+eyjsXGUoRSUZQHeQ5C4XLa0Y=
 github.com/imroc/req/v3 v3.57.0 h1:LMTUjNRUybUkTPn8oJDq8Kg3JRBOBTcnDhKu7mzupKI=
 github.com/imroc/req/v3 v3.57.0/go.mod h1:JL62ey1nvSLq81HORNcosvlf7SxZStONNqOprg0Pz00=
-github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
-github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
 github.com/jackc/pgpassfile v1.0.0 h1:/6Hmqy13Ss2zCq62VdNG8tM1wchn8zjSGOBJ6icpsIM=
 github.com/jackc/pgpassfile v1.0.0/go.mod h1:CEx0iS5ambNFdcRtxPj5JhEz+xB6uRky5eyVu/W2HEg=
 github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 h1:iCEnooe7UlwOQYpKFhBabPMi4aNAfoODPEFNiAnClxo=
@@ -286,10 +281,6 @@ github.com/pelletier/go-toml/v2 v2.2.2 h1:aYUidT7k73Pcl9nb2gScu7NSrKCSHIDE89b3+6
 github.com/pelletier/go-toml/v2 v2.2.2/go.mod h1:1t835xjRzz80PqgE6HHgN2JOsmgYu/h4qDAS4n929Rs=
 github.com/pkg/errors v0.9.1 h1:FEBLx1zS214owpjy7qsBeixbURkuhQAwrK5UwLGTwt4=
 github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
-github.com/pkoukk/tiktoken-go v0.1.8 h1:85ENo+3FpWgAACBaEUVp+lctuTcYUO7BtmfhlN/QTRo=
-github.com/pkoukk/tiktoken-go v0.1.8/go.mod h1:9NiV+i9mJKGj1rYOT+njbv+ZwA/zJxYdewGl6qVatpg=
-github.com/pkoukk/tiktoken-go-loader v0.0.2 h1:LUKws63GV3pVHwH1srkBplBv+7URgmOmhSkRxsIvsK4=
-github.com/pkoukk/tiktoken-go-loader v0.0.2/go.mod h1:4mIkYyZooFlnenDlormIo6cd5wrlUKNr97wp9nGgEKo=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 h1:Jamvg5psRIccs7FGNTlIRMkT8wgtp5eCXdBlqhYGL6U=
 github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=

backend/internal/domain/announcement.go

@@ -13,6 +13,11 @@ const (
 	AnnouncementStatusArchived = "archived"
 )
 
+const (
+	AnnouncementNotifyModeSilent = "silent"
+	AnnouncementNotifyModePopup  = "popup"
+)
+
 const (
 	AnnouncementConditionTypeSubscription = "subscription"
 	AnnouncementConditionTypeBalance      = "balance"
@@ -195,17 +200,18 @@ func (c AnnouncementCondition) validate() error {
 }
 
 type Announcement struct {
-	ID        int64
-	Title     string
-	Content   string
-	Status    string
-	Targeting AnnouncementTargeting
-	StartsAt  *time.Time
-	EndsAt    *time.Time
-	CreatedBy *int64
-	UpdatedBy *int64
-	CreatedAt time.Time
-	UpdatedAt time.Time
+	ID         int64
+	Title      string
+	Content    string
+	Status     string
+	NotifyMode string
+	Targeting  AnnouncementTargeting
+	StartsAt   *time.Time
+	EndsAt     *time.Time
+	CreatedBy  *int64
+	UpdatedBy  *int64
+	CreatedAt  time.Time
+	UpdatedAt  time.Time
 }
 
 func (a *Announcement) IsActiveAt(now time.Time) bool {

backend/internal/handler/admin/account_handler.go

@@ -122,7 +122,7 @@ type UpdateAccountRequest struct {
 	Priority                *int           `json:"priority"`
 	RateMultiplier          *float64       `json:"rate_multiplier"`
 	LoadFactor              *int           `json:"load_factor"`
-	Status                  string         `json:"status" binding:"omitempty,oneof=active inactive"`
+	Status                  string         `json:"status" binding:"omitempty,oneof=active inactive error"`
 	GroupIDs                *[]int64       `json:"group_ids"`
 	ExpiresAt               *int64         `json:"expires_at"`
 	AutoPauseOnExpired      *bool          `json:"auto_pause_on_expired"`
@@ -288,48 +288,32 @@ func (h *AccountHandler) List(c *gin.Context) {
 		}
 	}
 
-	// 窗口费用获取：lite 模式从快照缓存读取，非 lite 模式执行 PostgreSQL 查询后写入缓存
+	// 始终获取窗口费用（PostgreSQL 聚合查询）
 	if len(windowCostAccountIDs) > 0 {
-		if lite {
-			// lite 模式：尝试从快照缓存读取
-			cacheKey := buildWindowCostCacheKey(windowCostAccountIDs)
-			if cached, ok := accountWindowCostCache.Get(cacheKey); ok {
-				if costs, ok := cached.Payload.(map[int64]float64); ok {
-					windowCosts = costs
-				}
-			}
-			// 缓存未命中则 windowCosts 保持 nil（仅发生在服务刚启动时）
-		} else {
-			// 非 lite 模式：执行 PostgreSQL 聚合查询（高开销）
-			windowCosts = make(map[int64]float64)
-			var mu sync.Mutex
-			g, gctx := errgroup.WithContext(c.Request.Context())
-			g.SetLimit(10) // 限制并发数
+		windowCosts = make(map[int64]float64)
+		var mu sync.Mutex
+		g, gctx := errgroup.WithContext(c.Request.Context())
+		g.SetLimit(10) // 限制并发数
 
-			for i := range accounts {
-				acc := &accounts[i]
-				if !acc.IsAnthropicOAuthOrSetupToken() || acc.GetWindowCostLimit() <= 0 {
-					continue
-				}
-				accCopy := acc // 闭包捕获
-				g.Go(func() error {
-					// 使用统一的窗口开始时间计算逻辑（考虑窗口过期情况）
-					startTime := accCopy.GetCurrentWindowStartTime()
-					stats, err := h.accountUsageService.GetAccountWindowStats(gctx, accCopy.ID, startTime)
-					if err == nil && stats != nil {
-						mu.Lock()
-						windowCosts[accCopy.ID] = stats.StandardCost // 使用标准费用
-						mu.Unlock()
-					}
-					return nil // 不返回错误，允许部分失败
-				})
+		for i := range accounts {
+			acc := &accounts[i]
+			if !acc.IsAnthropicOAuthOrSetupToken() || acc.GetWindowCostLimit() <= 0 {
+				continue
 			}
-			_ = g.Wait()
-
-			// 查询完毕后写入快照缓存，供 lite 模式使用
-			cacheKey := buildWindowCostCacheKey(windowCostAccountIDs)
-			accountWindowCostCache.Set(cacheKey, windowCosts)
+			accCopy := acc // 闭包捕获
+			g.Go(func() error {
+				// 使用统一的窗口开始时间计算逻辑（考虑窗口过期情况）
+				startTime := accCopy.GetCurrentWindowStartTime()
+				stats, err := h.accountUsageService.GetAccountWindowStats(gctx, accCopy.ID, startTime)
+				if err == nil && stats != nil {
+					mu.Lock()
+					windowCosts[accCopy.ID] = stats.StandardCost // 使用标准费用
+					mu.Unlock()
+				}
+				return nil // 不返回错误，允许部分失败
+			})
 		}
+		_ = g.Wait()
 	}
 
 	// Build response with concurrency info

backend/internal/handler/admin/account_window_cost_cache.go

@@ -1,25 +0,0 @@
-package admin
-
-import (
-	"strconv"
-	"strings"
-	"time"
-)
-
-var accountWindowCostCache = newSnapshotCache(30 * time.Second)
-
-func buildWindowCostCacheKey(accountIDs []int64) string {
-	if len(accountIDs) == 0 {
-		return "accounts_window_cost_empty"
-	}
-	var b strings.Builder
-	b.Grow(len(accountIDs) * 6)
-	_, _ = b.WriteString("accounts_window_cost:")
-	for i, id := range accountIDs {
-		if i > 0 {
-			_ = b.WriteByte(',')
-		}
-		_, _ = b.WriteString(strconv.FormatInt(id, 10))
-	}
-	return b.String()
-}

backend/internal/handler/admin/announcement_handler.go

@@ -27,21 +27,23 @@ func NewAnnouncementHandler(announcementService *service.AnnouncementService) *A
 }
 
 type CreateAnnouncementRequest struct {
-	Title     string                        `json:"title" binding:"required"`
-	Content   string                        `json:"content" binding:"required"`
-	Status    string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
-	Targeting service.AnnouncementTargeting `json:"targeting"`
-	StartsAt  *int64                        `json:"starts_at"` // Unix seconds, 0/empty = immediate
-	EndsAt    *int64                        `json:"ends_at"`   // Unix seconds, 0/empty = never
+	Title      string                        `json:"title" binding:"required"`
+	Content    string                        `json:"content" binding:"required"`
+	Status     string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
+	NotifyMode string                        `json:"notify_mode" binding:"omitempty,oneof=silent popup"`
+	Targeting  service.AnnouncementTargeting `json:"targeting"`
+	StartsAt   *int64                        `json:"starts_at"` // Unix seconds, 0/empty = immediate
+	EndsAt     *int64                        `json:"ends_at"`   // Unix seconds, 0/empty = never
 }
 
 type UpdateAnnouncementRequest struct {
-	Title     *string                        `json:"title"`
-	Content   *string                        `json:"content"`
-	Status    *string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
-	Targeting *service.AnnouncementTargeting `json:"targeting"`
-	StartsAt  *int64                         `json:"starts_at"` // Unix seconds, 0 = clear
-	EndsAt    *int64                         `json:"ends_at"`   // Unix seconds, 0 = clear
+	Title      *string                        `json:"title"`
+	Content    *string                        `json:"content"`
+	Status     *string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
+	NotifyMode *string                        `json:"notify_mode" binding:"omitempty,oneof=silent popup"`
+	Targeting  *service.AnnouncementTargeting `json:"targeting"`
+	StartsAt   *int64                         `json:"starts_at"` // Unix seconds, 0 = clear
+	EndsAt     *int64                         `json:"ends_at"`   // Unix seconds, 0 = clear
 }
 
 // List handles listing announcements with filters
@@ -110,11 +112,12 @@ func (h *AnnouncementHandler) Create(c *gin.Context) {
 	}
 
 	input := &service.CreateAnnouncementInput{
-		Title:     req.Title,
-		Content:   req.Content,
-		Status:    req.Status,
-		Targeting: req.Targeting,
-		ActorID:   &subject.UserID,
+		Title:      req.Title,
+		Content:    req.Content,
+		Status:     req.Status,
+		NotifyMode: req.NotifyMode,
+		Targeting:  req.Targeting,
+		ActorID:    &subject.UserID,
 	}
 
 	if req.StartsAt != nil && *req.StartsAt > 0 {
@@ -157,11 +160,12 @@ func (h *AnnouncementHandler) Update(c *gin.Context) {
 	}
 
 	input := &service.UpdateAnnouncementInput{
-		Title:     req.Title,
-		Content:   req.Content,
-		Status:    req.Status,
-		Targeting: req.Targeting,
-		ActorID:   &subject.UserID,
+		Title:      req.Title,
+		Content:    req.Content,
+		Status:     req.Status,
+		NotifyMode: req.NotifyMode,
+		Targeting:  req.Targeting,
+		ActorID:    &subject.UserID,
 	}
 
 	if req.StartsAt != nil {

backend/internal/handler/admin/group_handler.go

@@ -53,6 +53,9 @@ type CreateGroupRequest struct {
 	SupportedModelScopes []string `json:"supported_model_scopes"`
 	// Sora 存储配额
 	SoraStorageQuotaBytes int64 `json:"sora_storage_quota_bytes"`
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch bool   `json:"allow_messages_dispatch"`
+	DefaultMappedModel    string `json:"default_mapped_model"`
 	// 从指定分组复制账号（创建后自动绑定）
 	CopyAccountsFromGroupIDs []int64 `json:"copy_accounts_from_group_ids"`
 }
@@ -88,6 +91,9 @@ type UpdateGroupRequest struct {
 	SupportedModelScopes *[]string `json:"supported_model_scopes"`
 	// Sora 存储配额
 	SoraStorageQuotaBytes *int64 `json:"sora_storage_quota_bytes"`
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch *bool   `json:"allow_messages_dispatch"`
+	DefaultMappedModel    *string `json:"default_mapped_model"`
 	// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
 	CopyAccountsFromGroupIDs []int64 `json:"copy_accounts_from_group_ids"`
 }
@@ -203,6 +209,8 @@ func (h *GroupHandler) Create(c *gin.Context) {
 		MCPXMLInject:                    req.MCPXMLInject,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           req.AllowMessagesDispatch,
+		DefaultMappedModel:              req.DefaultMappedModel,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})
 	if err != nil {
@@ -254,6 +262,8 @@ func (h *GroupHandler) Update(c *gin.Context) {
 		MCPXMLInject:                    req.MCPXMLInject,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           req.AllowMessagesDispatch,
+		DefaultMappedModel:              req.DefaultMappedModel,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})
 	if err != nil {

backend/internal/handler/dto/announcement.go

@@ -7,10 +7,11 @@ import (
 )
 
 type Announcement struct {
-	ID      int64  `json:"id"`
-	Title   string `json:"title"`
-	Content string `json:"content"`
-	Status  string `json:"status"`
+	ID         int64  `json:"id"`
+	Title      string `json:"title"`
+	Content    string `json:"content"`
+	Status     string `json:"status"`
+	NotifyMode string `json:"notify_mode"`
 
 	Targeting service.AnnouncementTargeting `json:"targeting"`
 
@@ -25,9 +26,10 @@ type Announcement struct {
 }
 
 type UserAnnouncement struct {
-	ID      int64  `json:"id"`
-	Title   string `json:"title"`
-	Content string `json:"content"`
+	ID         int64  `json:"id"`
+	Title      string `json:"title"`
+	Content    string `json:"content"`
+	NotifyMode string `json:"notify_mode"`
 
 	StartsAt *time.Time `json:"starts_at,omitempty"`
 	EndsAt   *time.Time `json:"ends_at,omitempty"`
@@ -43,17 +45,18 @@ func AnnouncementFromService(a *service.Announcement) *Announcement {
 		return nil
 	}
 	return &Announcement{
-		ID:        a.ID,
-		Title:     a.Title,
-		Content:   a.Content,
-		Status:    a.Status,
-		Targeting: a.Targeting,
-		StartsAt:  a.StartsAt,
-		EndsAt:    a.EndsAt,
-		CreatedBy: a.CreatedBy,
-		UpdatedBy: a.UpdatedBy,
-		CreatedAt: a.CreatedAt,
-		UpdatedAt: a.UpdatedAt,
+		ID:         a.ID,
+		Title:      a.Title,
+		Content:    a.Content,
+		Status:     a.Status,
+		NotifyMode: a.NotifyMode,
+		Targeting:  a.Targeting,
+		StartsAt:   a.StartsAt,
+		EndsAt:     a.EndsAt,
+		CreatedBy:  a.CreatedBy,
+		UpdatedBy:  a.UpdatedBy,
+		CreatedAt:  a.CreatedAt,
+		UpdatedAt:  a.UpdatedAt,
 	}
 }
 
@@ -62,13 +65,14 @@ func UserAnnouncementFromService(a *service.UserAnnouncement) *UserAnnouncement
 		return nil
 	}
 	return &UserAnnouncement{
-		ID:        a.Announcement.ID,
-		Title:     a.Announcement.Title,
-		Content:   a.Announcement.Content,
-		StartsAt:  a.Announcement.StartsAt,
-		EndsAt:    a.Announcement.EndsAt,
-		ReadAt:    a.ReadAt,
-		CreatedAt: a.Announcement.CreatedAt,
-		UpdatedAt: a.Announcement.UpdatedAt,
+		ID:         a.Announcement.ID,
+		Title:      a.Announcement.Title,
+		Content:    a.Announcement.Content,
+		NotifyMode: a.Announcement.NotifyMode,
+		StartsAt:   a.Announcement.StartsAt,
+		EndsAt:     a.Announcement.EndsAt,
+		ReadAt:     a.ReadAt,
+		CreatedAt:  a.Announcement.CreatedAt,
+		UpdatedAt:  a.Announcement.UpdatedAt,
 	}
 }

backend/internal/handler/dto/mappers.go

@@ -89,9 +89,9 @@ func APIKeyFromService(k *service.APIKey) *APIKey {
 		RateLimit5h:   k.RateLimit5h,
 		RateLimit1d:   k.RateLimit1d,
 		RateLimit7d:   k.RateLimit7d,
-		Usage5h:       k.Usage5h,
-		Usage1d:       k.Usage1d,
-		Usage7d:       k.Usage7d,
+		Usage5h:       k.EffectiveUsage5h(),
+		Usage1d:       k.EffectiveUsage1d(),
+		Usage7d:       k.EffectiveUsage7d(),
 		Window5hStart: k.Window5hStart,
 		Window1dStart: k.Window1dStart,
 		Window7dStart: k.Window7dStart,
@@ -125,8 +125,9 @@ func GroupFromServiceAdmin(g *service.Group) *AdminGroup {
 		Group:                groupFromServiceBase(g),
 		ModelRouting:         g.ModelRouting,
 		ModelRoutingEnabled:  g.ModelRoutingEnabled,
-		MCPXMLInject:         g.MCPXMLInject,
-		SupportedModelScopes: g.SupportedModelScopes,
+		MCPXMLInject:       g.MCPXMLInject,
+		DefaultMappedModel: g.DefaultMappedModel,
+		SupportedModelScopes:  g.SupportedModelScopes,
 		AccountCount:         g.AccountCount,
 		SortOrder:            g.SortOrder,
 	}
@@ -164,6 +165,7 @@ func groupFromServiceBase(g *service.Group) Group {
 		FallbackGroupID:                 g.FallbackGroupID,
 		FallbackGroupIDOnInvalidRequest: g.FallbackGroupIDOnInvalidRequest,
 		SoraStorageQuotaBytes:           g.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           g.AllowMessagesDispatch,
 		CreatedAt:                       g.CreatedAt,
 		UpdatedAt:                       g.UpdatedAt,
 	}

backend/internal/handler/dto/types.go

@@ -96,6 +96,9 @@ type Group struct {
 	// Sora 存储配额
 	SoraStorageQuotaBytes int64 `json:"sora_storage_quota_bytes"`
 
+	// OpenAI Messages 调度开关（用户侧需要此字段判断是否展示 Claude Code 教程）
+	AllowMessagesDispatch bool `json:"allow_messages_dispatch"`
+
 	CreatedAt time.Time `json:"created_at"`
 	UpdatedAt time.Time `json:"updated_at"`
 }
@@ -112,6 +115,9 @@ type AdminGroup struct {
 	// MCP XML 协议注入（仅 antigravity 平台使用）
 	MCPXMLInject bool `json:"mcp_xml_inject"`
 
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	DefaultMappedModel string `json:"default_mapped_model"`
+
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string       `json:"supported_model_scopes"`
 	AccountGroups        []AccountGroup `json:"account_groups,omitempty"`

backend/internal/handler/gateway_handler.go

@@ -971,7 +971,7 @@ func (h *GatewayHandler) usageQuotaLimited(c *gin.Context, ctx context.Context,
 		if err == nil && rateLimitData != nil {
 			var rateLimits []gin.H
 			if apiKey.RateLimit5h > 0 {
-				used := rateLimitData.Usage5h
+				used := rateLimitData.EffectiveUsage5h()
 				rateLimits = append(rateLimits, gin.H{
 					"window":       "5h",
 					"limit":        apiKey.RateLimit5h,
@@ -981,7 +981,7 @@ func (h *GatewayHandler) usageQuotaLimited(c *gin.Context, ctx context.Context,
 				})
 			}
 			if apiKey.RateLimit1d > 0 {
-				used := rateLimitData.Usage1d
+				used := rateLimitData.EffectiveUsage1d()
 				rateLimits = append(rateLimits, gin.H{
 					"window":       "1d",
 					"limit":        apiKey.RateLimit1d,
@@ -991,7 +991,7 @@ func (h *GatewayHandler) usageQuotaLimited(c *gin.Context, ctx context.Context,
 				})
 			}
 			if apiKey.RateLimit7d > 0 {
-				used := rateLimitData.Usage7d
+				used := rateLimitData.EffectiveUsage7d()
 				rateLimits = append(rateLimits, gin.H{
 					"window":       "7d",
 					"limit":        apiKey.RateLimit7d,

backend/internal/handler/openai_gateway_handler.go

@@ -118,6 +118,20 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, "", false, body)
+	sessionHashBody := body
+	if service.IsOpenAIResponsesCompactPathForTest(c) {
+		if compactSeed := strings.TrimSpace(gjson.GetBytes(body, "prompt_cache_key").String()); compactSeed != "" {
+			c.Set(service.OpenAICompactSessionSeedKeyForTest(), compactSeed)
+		}
+		normalizedCompactBody, normalizedCompact, compactErr := service.NormalizeOpenAICompactRequestBodyForTest(body)
+		if compactErr != nil {
+			h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to normalize compact request body")
+			return
+		}
+		if normalizedCompact {
+			body = normalizedCompactBody
+		}
+	}
 
 	// 校验请求体 JSON 合法性
 	if !gjson.ValidBytes(body) {
@@ -193,7 +207,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	}
 
 	// Generate session hash (header first; fallback to prompt_cache_key)
-	sessionHash := h.gatewayService.GenerateSessionHash(c, body)
+	sessionHash := h.gatewayService.GenerateSessionHash(c, sessionHashBody)
 
 	maxAccountSwitches := h.maxAccountSwitches
 	switchCount := 0
@@ -305,6 +319,9 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			return
 		}
 		if result != nil {
+			if account.Type == service.AccountTypeOAuth {
+				h.gatewayService.UpdateCodexUsageSnapshotFromHeaders(c.Request.Context(), account.ID, result.ResponseHeaders)
+			}
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
 		} else {
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, nil)
@@ -424,6 +441,316 @@ func (h *OpenAIGatewayHandler) logOpenAIRemoteCompactOutcome(c *gin.Context, sta
 	log.Warn("codex.remote_compact.failed")
 }
 
+// Messages handles Anthropic Messages API requests routed to OpenAI platform.
+// POST /v1/messages (when group platform is OpenAI)
+func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
+	streamStarted := false
+	defer h.recoverAnthropicMessagesPanic(c, &streamStarted)
+
+	requestStart := time.Now()
+
+	apiKey, ok := middleware2.GetAPIKeyFromContext(c)
+	if !ok {
+		h.anthropicErrorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
+		return
+	}
+
+	subject, ok := middleware2.GetAuthSubjectFromContext(c)
+	if !ok {
+		h.anthropicErrorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
+		return
+	}
+	reqLog := requestLogger(
+		c,
+		"handler.openai_gateway.messages",
+		zap.Int64("user_id", subject.UserID),
+		zap.Int64("api_key_id", apiKey.ID),
+		zap.Any("group_id", apiKey.GroupID),
+	)
+
+	// 检查分组是否允许 /v1/messages 调度
+	if apiKey.Group != nil && !apiKey.Group.AllowMessagesDispatch {
+		h.anthropicErrorResponse(c, http.StatusForbidden, "permission_error",
+			"This group does not allow /v1/messages dispatch")
+		return
+	}
+
+	if !h.ensureResponsesDependencies(c, reqLog) {
+		return
+	}
+
+	body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
+	if err != nil {
+		if maxErr, ok := extractMaxBytesError(err); ok {
+			h.anthropicErrorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
+			return
+		}
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
+		return
+	}
+	if len(body) == 0 {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
+		return
+	}
+
+	if !gjson.ValidBytes(body) {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return
+	}
+
+	modelResult := gjson.GetBytes(body, "model")
+	if !modelResult.Exists() || modelResult.Type != gjson.String || modelResult.String() == "" {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return
+	}
+	reqModel := modelResult.String()
+	reqStream := gjson.GetBytes(body, "stream").Bool()
+
+	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
+
+	setOpsRequestContext(c, reqModel, reqStream, body)
+
+	// 绑定错误透传服务，允许 service 层在非 failover 错误场景复用规则。
+	if h.errorPassthroughService != nil {
+		service.BindErrorPassthroughService(c, h.errorPassthroughService)
+	}
+
+	subscription, _ := middleware2.GetSubscriptionFromContext(c)
+
+	service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
+	routingStart := time.Now()
+
+	userReleaseFunc, acquired := h.acquireResponsesUserSlot(c, subject.UserID, subject.Concurrency, reqStream, &streamStarted, reqLog)
+	if !acquired {
+		return
+	}
+	if userReleaseFunc != nil {
+		defer userReleaseFunc()
+	}
+
+	if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription); err != nil {
+		reqLog.Info("openai_messages.billing_eligibility_check_failed", zap.Error(err))
+		status, code, message := billingErrorDetails(err)
+		h.anthropicStreamingAwareError(c, status, code, message, streamStarted)
+		return
+	}
+
+	sessionHash := h.gatewayService.GenerateSessionHash(c, body)
+	promptCacheKey := h.gatewayService.ExtractSessionID(c, body)
+
+	maxAccountSwitches := h.maxAccountSwitches
+	switchCount := 0
+	failedAccountIDs := make(map[int64]struct{})
+	var lastFailoverErr *service.UpstreamFailoverError
+
+	for {
+		// 清除上一次迭代的降级模型标记，避免残留影响本次迭代
+		c.Set("openai_messages_fallback_model", "")
+		reqLog.Debug("openai_messages.account_selecting", zap.Int("excluded_account_count", len(failedAccountIDs)))
+		selection, scheduleDecision, err := h.gatewayService.SelectAccountWithScheduler(
+			c.Request.Context(),
+			apiKey.GroupID,
+			"", // no previous_response_id
+			sessionHash,
+			reqModel,
+			failedAccountIDs,
+			service.OpenAIUpstreamTransportAny,
+		)
+		if err != nil {
+			reqLog.Warn("openai_messages.account_select_failed",
+				zap.Error(err),
+				zap.Int("excluded_account_count", len(failedAccountIDs)),
+			)
+			// 首次调度失败 + 有默认映射模型 → 用默认模型重试
+			if len(failedAccountIDs) == 0 {
+				defaultModel := ""
+				if apiKey.Group != nil {
+					defaultModel = apiKey.Group.DefaultMappedModel
+				}
+				if defaultModel != "" && defaultModel != reqModel {
+					reqLog.Info("openai_messages.fallback_to_default_model",
+						zap.String("default_mapped_model", defaultModel),
+					)
+					selection, scheduleDecision, err = h.gatewayService.SelectAccountWithScheduler(
+						c.Request.Context(),
+						apiKey.GroupID,
+						"",
+						sessionHash,
+						defaultModel,
+						failedAccountIDs,
+						service.OpenAIUpstreamTransportAny,
+					)
+					if err == nil && selection != nil {
+						c.Set("openai_messages_fallback_model", defaultModel)
+					}
+				}
+				if err != nil {
+					h.anthropicStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "Service temporarily unavailable", streamStarted)
+					return
+				}
+			} else {
+				if lastFailoverErr != nil {
+					h.handleAnthropicFailoverExhausted(c, lastFailoverErr, streamStarted)
+				} else {
+					h.anthropicStreamingAwareError(c, http.StatusBadGateway, "api_error", "Upstream request failed", streamStarted)
+				}
+				return
+			}
+		}
+		if selection == nil || selection.Account == nil {
+			h.anthropicStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+			return
+		}
+		account := selection.Account
+		reqLog.Debug("openai_messages.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
+		_ = scheduleDecision
+		setOpsSelectedAccount(c, account.ID, account.Platform)
+
+		accountReleaseFunc, acquired := h.acquireResponsesAccountSlot(c, apiKey.GroupID, sessionHash, selection, reqStream, &streamStarted, reqLog)
+		if !acquired {
+			return
+		}
+
+		service.SetOpsLatencyMs(c, service.OpsRoutingLatencyMsKey, time.Since(routingStart).Milliseconds())
+		forwardStart := time.Now()
+
+		defaultMappedModel := ""
+		if apiKey.Group != nil {
+			defaultMappedModel = apiKey.Group.DefaultMappedModel
+		}
+		// 如果使用了降级模型调度，强制使用降级模型
+		if fallbackModel := c.GetString("openai_messages_fallback_model"); fallbackModel != "" {
+			defaultMappedModel = fallbackModel
+		}
+		result, err := h.gatewayService.ForwardAsAnthropic(c.Request.Context(), c, account, body, promptCacheKey, defaultMappedModel)
+
+		forwardDurationMs := time.Since(forwardStart).Milliseconds()
+		if accountReleaseFunc != nil {
+			accountReleaseFunc()
+		}
+		upstreamLatencyMs, _ := getContextInt64(c, service.OpsUpstreamLatencyMsKey)
+		responseLatencyMs := forwardDurationMs
+		if upstreamLatencyMs > 0 && forwardDurationMs > upstreamLatencyMs {
+			responseLatencyMs = forwardDurationMs - upstreamLatencyMs
+		}
+		service.SetOpsLatencyMs(c, service.OpsResponseLatencyMsKey, responseLatencyMs)
+		if err == nil && result != nil && result.FirstTokenMs != nil {
+			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
+		}
+		if err != nil {
+			var failoverErr *service.UpstreamFailoverError
+			if errors.As(err, &failoverErr) {
+				h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+				h.gatewayService.RecordOpenAIAccountSwitch()
+				failedAccountIDs[account.ID] = struct{}{}
+				lastFailoverErr = failoverErr
+				if switchCount >= maxAccountSwitches {
+					h.handleAnthropicFailoverExhausted(c, failoverErr, streamStarted)
+					return
+				}
+				switchCount++
+				reqLog.Warn("openai_messages.upstream_failover_switching",
+					zap.Int64("account_id", account.ID),
+					zap.Int("upstream_status", failoverErr.StatusCode),
+					zap.Int("switch_count", switchCount),
+					zap.Int("max_switches", maxAccountSwitches),
+				)
+				continue
+			}
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+			wroteFallback := h.ensureAnthropicErrorResponse(c, streamStarted)
+			reqLog.Warn("openai_messages.forward_failed",
+				zap.Int64("account_id", account.ID),
+				zap.Bool("fallback_error_response_written", wroteFallback),
+				zap.Error(err),
+			)
+			return
+		}
+		if result != nil {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
+		} else {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, nil)
+		}
+
+		userAgent := c.GetHeader("User-Agent")
+		clientIP := ip.GetClientIP(c)
+
+		h.submitUsageRecordTask(func(ctx context.Context) {
+			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+				Result:        result,
+				APIKey:        apiKey,
+				User:          apiKey.User,
+				Account:       account,
+				Subscription:  subscription,
+				UserAgent:     userAgent,
+				IPAddress:     clientIP,
+				APIKeyService: h.apiKeyService,
+			}); err != nil {
+				logger.L().With(
+					zap.String("component", "handler.openai_gateway.messages"),
+					zap.Int64("user_id", subject.UserID),
+					zap.Int64("api_key_id", apiKey.ID),
+					zap.Any("group_id", apiKey.GroupID),
+					zap.String("model", reqModel),
+					zap.Int64("account_id", account.ID),
+				).Error("openai_messages.record_usage_failed", zap.Error(err))
+			}
+		})
+		reqLog.Debug("openai_messages.request_completed",
+			zap.Int64("account_id", account.ID),
+			zap.Int("switch_count", switchCount),
+		)
+		return
+	}
+}
+
+// anthropicErrorResponse writes an error in Anthropic Messages API format.
+func (h *OpenAIGatewayHandler) anthropicErrorResponse(c *gin.Context, status int, errType, message string) {
+	c.JSON(status, gin.H{
+		"type": "error",
+		"error": gin.H{
+			"type":    errType,
+			"message": message,
+		},
+	})
+}
+
+// anthropicStreamingAwareError handles errors that may occur during streaming,
+// using Anthropic SSE error format.
+func (h *OpenAIGatewayHandler) anthropicStreamingAwareError(c *gin.Context, status int, errType, message string, streamStarted bool) {
+	if streamStarted {
+		flusher, ok := c.Writer.(http.Flusher)
+		if ok {
+			errPayload, _ := json.Marshal(gin.H{
+				"type": "error",
+				"error": gin.H{
+					"type":    errType,
+					"message": message,
+				},
+			})
+			fmt.Fprintf(c.Writer, "event: error\ndata: %s\n\n", errPayload) //nolint:errcheck
+			flusher.Flush()
+		}
+		return
+	}
+	h.anthropicErrorResponse(c, status, errType, message)
+}
+
+// handleAnthropicFailoverExhausted maps upstream failover errors to Anthropic format.
+func (h *OpenAIGatewayHandler) handleAnthropicFailoverExhausted(c *gin.Context, failoverErr *service.UpstreamFailoverError, streamStarted bool) {
+	status, errType, errMsg := h.mapUpstreamError(failoverErr.StatusCode)
+	h.anthropicStreamingAwareError(c, status, errType, errMsg, streamStarted)
+}
+
+// ensureAnthropicErrorResponse writes a fallback Anthropic error if no response was written.
+func (h *OpenAIGatewayHandler) ensureAnthropicErrorResponse(c *gin.Context, streamStarted bool) bool {
+	if c == nil || c.Writer == nil || c.Writer.Written() {
+		return false
+	}
+	h.anthropicStreamingAwareError(c, http.StatusBadGateway, "api_error", "Upstream request failed", streamStarted)
+	return true
+}
+
 func (h *OpenAIGatewayHandler) validateFunctionCallOutputRequest(c *gin.Context, body []byte, reqLog *zap.Logger) bool {
 	if !gjson.GetBytes(body, `input.#(type=="function_call_output")`).Exists() {
 		return true
@@ -840,6 +1167,9 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 			if turnErr != nil || result == nil {
 				return
 			}
+			if account.Type == service.AccountTypeOAuth {
+				h.gatewayService.UpdateCodexUsageSnapshotFromHeaders(ctx, account.ID, result.ResponseHeaders)
+			}
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
 			h.submitUsageRecordTask(func(taskCtx context.Context) {
 				if err := h.gatewayService.RecordUsage(taskCtx, &service.OpenAIRecordUsageInput{
@@ -901,6 +1231,26 @@ func (h *OpenAIGatewayHandler) recoverResponsesPanic(c *gin.Context, streamStart
 	)
 }
 
+// recoverAnthropicMessagesPanic recovers from panics in the Anthropic Messages
+// handler and returns an Anthropic-formatted error response.
+func (h *OpenAIGatewayHandler) recoverAnthropicMessagesPanic(c *gin.Context, streamStarted *bool) {
+	recovered := recover()
+	if recovered == nil {
+		return
+	}
+
+	started := streamStarted != nil && *streamStarted
+	requestLogger(c, "handler.openai_gateway.messages").Error(
+		"openai.messages_panic_recovered",
+		zap.Bool("stream_started", started),
+		zap.Any("panic", recovered),
+		zap.ByteString("stack", debug.Stack()),
+	)
+	if !started {
+		h.anthropicErrorResponse(c, http.StatusInternalServerError, "api_error", "Internal server error")
+	}
+}
+
 func (h *OpenAIGatewayHandler) ensureResponsesDependencies(c *gin.Context, reqLog *zap.Logger) bool {
 	missing := h.missingResponsesDependencies()
 	if len(missing) == 0 {

backend/internal/pkg/antigravity/stream_transformer.go

@@ -119,23 +119,33 @@ func (p *StreamingProcessor) ProcessLine(line string) []byte {
 	return result.Bytes()
 }
 
-// Finish 结束处理，返回最终事件和用量
+// Finish 结束处理，返回最终事件和用量。
+// 若整个流未收到任何可解析的上游数据（messageStartSent == false），
+// 则不补发任何结束事件，防止客户端收到没有 message_start 的残缺流。
 func (p *StreamingProcessor) Finish() ([]byte, *ClaudeUsage) {
-	var result bytes.Buffer
-
-	if !p.messageStopSent {
-		_, _ = result.Write(p.emitFinish(""))
-	}
-
 	usage := &ClaudeUsage{
 		InputTokens:          p.inputTokens,
 		OutputTokens:         p.outputTokens,
 		CacheReadInputTokens: p.cacheReadTokens,
 	}
 
+	if !p.messageStartSent {
+		return nil, usage
+	}
+
+	var result bytes.Buffer
+	if !p.messageStopSent {
+		_, _ = result.Write(p.emitFinish(""))
+	}
+
 	return result.Bytes(), usage
 }
 
+// MessageStartSent 报告流中是否已发出过 message_start 事件（即是否收到过有效的上游数据）
+func (p *StreamingProcessor) MessageStartSent() bool {
+	return p.messageStartSent
+}
+
 // emitMessageStart 发送 message_start 事件
 func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte {
 	if p.messageStartSent {

backend/internal/pkg/apicompat/anthropic_responses_test.go

@@ -0,0 +1,735 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// ---------------------------------------------------------------------------
+// AnthropicToResponses tests
+// ---------------------------------------------------------------------------
+
+func TestAnthropicToResponses_BasicText(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Stream:    true,
+		Messages: []AnthropicMessage{
+			{Role: "user", Content: json.RawMessage(`"Hello"`)},
+		},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-5.2", resp.Model)
+	assert.True(t, resp.Stream)
+	assert.Equal(t, 1024, *resp.MaxOutputTokens)
+	assert.False(t, *resp.Store)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 1)
+	assert.Equal(t, "user", items[0].Role)
+}
+
+func TestAnthropicToResponses_SystemPrompt(t *testing.T) {
+	t.Run("string", func(t *testing.T) {
+		req := &AnthropicRequest{
+			Model:     "gpt-5.2",
+			MaxTokens: 100,
+			System:    json.RawMessage(`"You are helpful."`),
+			Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+		}
+		resp, err := AnthropicToResponses(req)
+		require.NoError(t, err)
+
+		var items []ResponsesInputItem
+		require.NoError(t, json.Unmarshal(resp.Input, &items))
+		require.Len(t, items, 2)
+		assert.Equal(t, "system", items[0].Role)
+	})
+
+	t.Run("array", func(t *testing.T) {
+		req := &AnthropicRequest{
+			Model:     "gpt-5.2",
+			MaxTokens: 100,
+			System:    json.RawMessage(`[{"type":"text","text":"Part 1"},{"type":"text","text":"Part 2"}]`),
+			Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+		}
+		resp, err := AnthropicToResponses(req)
+		require.NoError(t, err)
+
+		var items []ResponsesInputItem
+		require.NoError(t, json.Unmarshal(resp.Input, &items))
+		require.Len(t, items, 2)
+		assert.Equal(t, "system", items[0].Role)
+		// System text should be joined with double newline.
+		var text string
+		require.NoError(t, json.Unmarshal(items[0].Content, &text))
+		assert.Equal(t, "Part 1\n\nPart 2", text)
+	})
+}
+
+func TestAnthropicToResponses_ToolUse(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages: []AnthropicMessage{
+			{Role: "user", Content: json.RawMessage(`"What is the weather?"`)},
+			{Role: "assistant", Content: json.RawMessage(`[{"type":"text","text":"Let me check."},{"type":"tool_use","id":"call_1","name":"get_weather","input":{"city":"NYC"}}]`)},
+			{Role: "user", Content: json.RawMessage(`[{"type":"tool_result","tool_use_id":"call_1","content":"Sunny, 72°F"}]`)},
+		},
+		Tools: []AnthropicTool{
+			{Name: "get_weather", Description: "Get weather", InputSchema: json.RawMessage(`{"type":"object","properties":{"city":{"type":"string"}}}`)},
+		},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	// Check tools
+	require.Len(t, resp.Tools, 1)
+	assert.Equal(t, "function", resp.Tools[0].Type)
+	assert.Equal(t, "get_weather", resp.Tools[0].Name)
+
+	// Check input items
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	// user + assistant + function_call + function_call_output = 4
+	require.Len(t, items, 4)
+
+	assert.Equal(t, "user", items[0].Role)
+	assert.Equal(t, "assistant", items[1].Role)
+	assert.Equal(t, "function_call", items[2].Type)
+	assert.Equal(t, "fc_call_1", items[2].CallID)
+	assert.Equal(t, "function_call_output", items[3].Type)
+	assert.Equal(t, "fc_call_1", items[3].CallID)
+	assert.Equal(t, "Sunny, 72°F", items[3].Output)
+}
+
+func TestAnthropicToResponses_ThinkingIgnored(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages: []AnthropicMessage{
+			{Role: "user", Content: json.RawMessage(`"Hello"`)},
+			{Role: "assistant", Content: json.RawMessage(`[{"type":"thinking","thinking":"deep thought"},{"type":"text","text":"Hi!"}]`)},
+			{Role: "user", Content: json.RawMessage(`"More"`)},
+		},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	// user + assistant(text only, thinking ignored) + user = 3
+	require.Len(t, items, 3)
+	assert.Equal(t, "assistant", items[1].Role)
+	// Assistant content should only have text, not thinking.
+	var parts []ResponsesContentPart
+	require.NoError(t, json.Unmarshal(items[1].Content, &parts))
+	require.Len(t, parts, 1)
+	assert.Equal(t, "output_text", parts[0].Type)
+	assert.Equal(t, "Hi!", parts[0].Text)
+}
+
+func TestAnthropicToResponses_MaxTokensFloor(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 10, // below minMaxOutputTokens (128)
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	assert.Equal(t, 128, *resp.MaxOutputTokens)
+}
+
+// ---------------------------------------------------------------------------
+// ResponsesToAnthropic (non-streaming) tests
+// ---------------------------------------------------------------------------
+
+func TestResponsesToAnthropic_TextOnly(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_123",
+		Model:  "gpt-5.2",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "Hello there!"},
+				},
+			},
+		},
+		Usage: &ResponsesUsage{InputTokens: 10, OutputTokens: 5, TotalTokens: 15},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	assert.Equal(t, "resp_123", anth.ID)
+	assert.Equal(t, "claude-opus-4-6", anth.Model)
+	assert.Equal(t, "end_turn", anth.StopReason)
+	require.Len(t, anth.Content, 1)
+	assert.Equal(t, "text", anth.Content[0].Type)
+	assert.Equal(t, "Hello there!", anth.Content[0].Text)
+	assert.Equal(t, 10, anth.Usage.InputTokens)
+	assert.Equal(t, 5, anth.Usage.OutputTokens)
+}
+
+func TestResponsesToAnthropic_ToolUse(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_456",
+		Model:  "gpt-5.2",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "Let me check."},
+				},
+			},
+			{
+				Type:      "function_call",
+				CallID:    "call_1",
+				Name:      "get_weather",
+				Arguments: `{"city":"NYC"}`,
+			},
+		},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	assert.Equal(t, "tool_use", anth.StopReason)
+	require.Len(t, anth.Content, 2)
+	assert.Equal(t, "text", anth.Content[0].Type)
+	assert.Equal(t, "tool_use", anth.Content[1].Type)
+	assert.Equal(t, "call_1", anth.Content[1].ID)
+	assert.Equal(t, "get_weather", anth.Content[1].Name)
+}
+
+func TestResponsesToAnthropic_Reasoning(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_789",
+		Model:  "gpt-5.2",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type: "reasoning",
+				Summary: []ResponsesSummary{
+					{Type: "summary_text", Text: "Thinking about the answer..."},
+				},
+			},
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "42"},
+				},
+			},
+		},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	require.Len(t, anth.Content, 2)
+	assert.Equal(t, "thinking", anth.Content[0].Type)
+	assert.Equal(t, "Thinking about the answer...", anth.Content[0].Thinking)
+	assert.Equal(t, "text", anth.Content[1].Type)
+	assert.Equal(t, "42", anth.Content[1].Text)
+}
+
+func TestResponsesToAnthropic_Incomplete(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_inc",
+		Model:  "gpt-5.2",
+		Status: "incomplete",
+		IncompleteDetails: &ResponsesIncompleteDetails{
+			Reason: "max_output_tokens",
+		},
+		Output: []ResponsesOutput{
+			{
+				Type:    "message",
+				Content: []ResponsesContentPart{{Type: "output_text", Text: "Partial..."}},
+			},
+		},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	assert.Equal(t, "max_tokens", anth.StopReason)
+}
+
+func TestResponsesToAnthropic_EmptyOutput(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_empty",
+		Model:  "gpt-5.2",
+		Status: "completed",
+		Output: []ResponsesOutput{},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	require.Len(t, anth.Content, 1)
+	assert.Equal(t, "text", anth.Content[0].Type)
+	assert.Equal(t, "", anth.Content[0].Text)
+}
+
+// ---------------------------------------------------------------------------
+// Streaming: ResponsesEventToAnthropicEvents tests
+// ---------------------------------------------------------------------------
+
+func TestStreamingTextOnly(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	// 1. response.created
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.created",
+		Response: &ResponsesResponse{
+			ID:    "resp_1",
+			Model: "gpt-5.2",
+		},
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "message_start", events[0].Type)
+
+	// 2. output_item.added (message)
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:        "response.output_item.added",
+		OutputIndex: 0,
+		Item:        &ResponsesOutput{Type: "message"},
+	}, state)
+	assert.Len(t, events, 0) // message item doesn't emit events
+
+	// 3. text delta
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "Hello",
+	}, state)
+	require.Len(t, events, 2) // content_block_start + content_block_delta
+	assert.Equal(t, "content_block_start", events[0].Type)
+	assert.Equal(t, "text", events[0].ContentBlock.Type)
+	assert.Equal(t, "content_block_delta", events[1].Type)
+	assert.Equal(t, "text_delta", events[1].Delta.Type)
+	assert.Equal(t, "Hello", events[1].Delta.Text)
+
+	// 4. more text
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: " world",
+	}, state)
+	require.Len(t, events, 1) // only delta, no new block start
+	assert.Equal(t, "content_block_delta", events[0].Type)
+
+	// 5. text done
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.output_text.done",
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_stop", events[0].Type)
+
+	// 6. completed
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage:  &ResponsesUsage{InputTokens: 10, OutputTokens: 5},
+		},
+	}, state)
+	require.Len(t, events, 2) // message_delta + message_stop
+	assert.Equal(t, "message_delta", events[0].Type)
+	assert.Equal(t, "end_turn", events[0].Delta.StopReason)
+	assert.Equal(t, 10, events[0].Usage.InputTokens)
+	assert.Equal(t, 5, events[0].Usage.OutputTokens)
+	assert.Equal(t, "message_stop", events[1].Type)
+}
+
+func TestStreamingToolCall(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	// 1. response.created
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_2", Model: "gpt-5.2"},
+	}, state)
+
+	// 2. function_call added
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:        "response.output_item.added",
+		OutputIndex: 0,
+		Item:        &ResponsesOutput{Type: "function_call", CallID: "call_1", Name: "get_weather"},
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_start", events[0].Type)
+	assert.Equal(t, "tool_use", events[0].ContentBlock.Type)
+	assert.Equal(t, "call_1", events[0].ContentBlock.ID)
+
+	// 3. arguments delta
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:        "response.function_call_arguments.delta",
+		OutputIndex: 0,
+		Delta:       `{"city":`,
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_delta", events[0].Type)
+	assert.Equal(t, "input_json_delta", events[0].Delta.Type)
+	assert.Equal(t, `{"city":`, events[0].Delta.PartialJSON)
+
+	// 4. arguments done
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.function_call_arguments.done",
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_stop", events[0].Type)
+
+	// 5. completed with tool_calls
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage:  &ResponsesUsage{InputTokens: 20, OutputTokens: 10},
+		},
+	}, state)
+	require.Len(t, events, 2)
+	assert.Equal(t, "tool_use", events[0].Delta.StopReason)
+}
+
+func TestStreamingReasoning(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_3", Model: "gpt-5.2"},
+	}, state)
+
+	// reasoning item added
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:        "response.output_item.added",
+		OutputIndex: 0,
+		Item:        &ResponsesOutput{Type: "reasoning"},
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_start", events[0].Type)
+	assert.Equal(t, "thinking", events[0].ContentBlock.Type)
+
+	// reasoning text delta
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:        "response.reasoning_summary_text.delta",
+		OutputIndex: 0,
+		Delta:       "Let me think...",
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_delta", events[0].Type)
+	assert.Equal(t, "thinking_delta", events[0].Delta.Type)
+	assert.Equal(t, "Let me think...", events[0].Delta.Thinking)
+
+	// reasoning done
+	events = ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.reasoning_summary_text.done",
+	}, state)
+	require.Len(t, events, 1)
+	assert.Equal(t, "content_block_stop", events[0].Type)
+}
+
+func TestStreamingIncomplete(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_4", Model: "gpt-5.2"},
+	}, state)
+
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "Partial output...",
+	}, state)
+
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.incomplete",
+		Response: &ResponsesResponse{
+			Status:            "incomplete",
+			IncompleteDetails: &ResponsesIncompleteDetails{Reason: "max_output_tokens"},
+			Usage:             &ResponsesUsage{InputTokens: 100, OutputTokens: 4096},
+		},
+	}, state)
+
+	// Should close the text block + message_delta + message_stop
+	require.Len(t, events, 3)
+	assert.Equal(t, "content_block_stop", events[0].Type)
+	assert.Equal(t, "message_delta", events[1].Type)
+	assert.Equal(t, "max_tokens", events[1].Delta.StopReason)
+	assert.Equal(t, "message_stop", events[2].Type)
+}
+
+func TestFinalizeStream_NeverStarted(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+	events := FinalizeResponsesAnthropicStream(state)
+	assert.Nil(t, events)
+}
+
+func TestFinalizeStream_AlreadyCompleted(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+	state.MessageStartSent = true
+	state.MessageStopSent = true
+	events := FinalizeResponsesAnthropicStream(state)
+	assert.Nil(t, events)
+}
+
+func TestFinalizeStream_AbnormalTermination(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	// Simulate a stream that started but never completed
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_5", Model: "gpt-5.2"},
+	}, state)
+
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "Interrupted...",
+	}, state)
+
+	// Stream ends without response.completed
+	events := FinalizeResponsesAnthropicStream(state)
+	require.Len(t, events, 3) // content_block_stop + message_delta + message_stop
+	assert.Equal(t, "content_block_stop", events[0].Type)
+	assert.Equal(t, "message_delta", events[1].Type)
+	assert.Equal(t, "end_turn", events[1].Delta.StopReason)
+	assert.Equal(t, "message_stop", events[2].Type)
+}
+
+func TestStreamingEmptyResponse(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_6", Model: "gpt-5.2"},
+	}, state)
+
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage:  &ResponsesUsage{InputTokens: 5, OutputTokens: 0},
+		},
+	}, state)
+
+	require.Len(t, events, 2) // message_delta + message_stop
+	assert.Equal(t, "message_delta", events[0].Type)
+	assert.Equal(t, "end_turn", events[0].Delta.StopReason)
+}
+
+func TestResponsesAnthropicEventToSSE(t *testing.T) {
+	evt := AnthropicStreamEvent{
+		Type: "message_start",
+		Message: &AnthropicResponse{
+			ID:   "resp_1",
+			Type: "message",
+			Role: "assistant",
+		},
+	}
+	sse, err := ResponsesAnthropicEventToSSE(evt)
+	require.NoError(t, err)
+	assert.Contains(t, sse, "event: message_start\n")
+	assert.Contains(t, sse, "data: ")
+	assert.Contains(t, sse, `"resp_1"`)
+}
+
+// ---------------------------------------------------------------------------
+// response.failed tests
+// ---------------------------------------------------------------------------
+
+func TestStreamingFailed(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	// 1. response.created
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_fail_1", Model: "gpt-5.2"},
+	}, state)
+
+	// 2. Some text output before failure
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "Partial output before failure",
+	}, state)
+
+	// 3. response.failed
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.failed",
+		Response: &ResponsesResponse{
+			Status: "failed",
+			Error:  &ResponsesError{Code: "server_error", Message: "Internal error"},
+			Usage:  &ResponsesUsage{InputTokens: 50, OutputTokens: 10},
+		},
+	}, state)
+
+	// Should close text block + message_delta + message_stop
+	require.Len(t, events, 3)
+	assert.Equal(t, "content_block_stop", events[0].Type)
+	assert.Equal(t, "message_delta", events[1].Type)
+	assert.Equal(t, "end_turn", events[1].Delta.StopReason)
+	assert.Equal(t, 50, events[1].Usage.InputTokens)
+	assert.Equal(t, 10, events[1].Usage.OutputTokens)
+	assert.Equal(t, "message_stop", events[2].Type)
+}
+
+func TestStreamingFailedNoOutput(t *testing.T) {
+	state := NewResponsesEventToAnthropicState()
+
+	// 1. response.created
+	ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_fail_2", Model: "gpt-5.2"},
+	}, state)
+
+	// 2. response.failed with no prior output
+	events := ResponsesEventToAnthropicEvents(&ResponsesStreamEvent{
+		Type: "response.failed",
+		Response: &ResponsesResponse{
+			Status: "failed",
+			Error:  &ResponsesError{Code: "rate_limit_error", Message: "Too many requests"},
+			Usage:  &ResponsesUsage{InputTokens: 20, OutputTokens: 0},
+		},
+	}, state)
+
+	// Should emit message_delta + message_stop (no block to close)
+	require.Len(t, events, 2)
+	assert.Equal(t, "message_delta", events[0].Type)
+	assert.Equal(t, "end_turn", events[0].Delta.StopReason)
+	assert.Equal(t, "message_stop", events[1].Type)
+}
+
+func TestResponsesToAnthropic_Failed(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_fail_3",
+		Model:  "gpt-5.2",
+		Status: "failed",
+		Error:  &ResponsesError{Code: "server_error", Message: "Something went wrong"},
+		Output: []ResponsesOutput{},
+		Usage:  &ResponsesUsage{InputTokens: 30, OutputTokens: 0},
+	}
+
+	anth := ResponsesToAnthropic(resp, "claude-opus-4-6")
+	// Failed status defaults to "end_turn" stop reason
+	assert.Equal(t, "end_turn", anth.StopReason)
+	// Should have at least an empty text block
+	require.Len(t, anth.Content, 1)
+	assert.Equal(t, "text", anth.Content[0].Type)
+}
+
+// ---------------------------------------------------------------------------
+// thinking → reasoning conversion tests
+// ---------------------------------------------------------------------------
+
+func TestAnthropicToResponses_ThinkingEnabled(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		Thinking:  &AnthropicThinking{Type: "enabled", BudgetTokens: 10000},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "high", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+	assert.Contains(t, resp.Include, "reasoning.encrypted_content")
+	assert.NotContains(t, resp.Include, "reasoning.summary")
+}
+
+func TestAnthropicToResponses_ThinkingAdaptive(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		Thinking:  &AnthropicThinking{Type: "adaptive", BudgetTokens: 5000},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "medium", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+	assert.NotContains(t, resp.Include, "reasoning.summary")
+}
+
+func TestAnthropicToResponses_ThinkingDisabled(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		Thinking:  &AnthropicThinking{Type: "disabled"},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	assert.Nil(t, resp.Reasoning)
+	assert.NotContains(t, resp.Include, "reasoning.summary")
+}
+
+func TestAnthropicToResponses_NoThinking(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	assert.Nil(t, resp.Reasoning)
+}
+
+// ---------------------------------------------------------------------------
+// tool_choice conversion tests
+// ---------------------------------------------------------------------------
+
+func TestAnthropicToResponses_ToolChoiceAuto(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:      "gpt-5.2",
+		MaxTokens:  1024,
+		Messages:   []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		ToolChoice: json.RawMessage(`{"type":"auto"}`),
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	var tc string
+	require.NoError(t, json.Unmarshal(resp.ToolChoice, &tc))
+	assert.Equal(t, "auto", tc)
+}
+
+func TestAnthropicToResponses_ToolChoiceAny(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:      "gpt-5.2",
+		MaxTokens:  1024,
+		Messages:   []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		ToolChoice: json.RawMessage(`{"type":"any"}`),
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	var tc string
+	require.NoError(t, json.Unmarshal(resp.ToolChoice, &tc))
+	assert.Equal(t, "required", tc)
+}
+
+func TestAnthropicToResponses_ToolChoiceSpecific(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:      "gpt-5.2",
+		MaxTokens:  1024,
+		Messages:   []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		ToolChoice: json.RawMessage(`{"type":"tool","name":"get_weather"}`),
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	var tc map[string]any
+	require.NoError(t, json.Unmarshal(resp.ToolChoice, &tc))
+	assert.Equal(t, "function", tc["type"])
+	fn, ok := tc["function"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, "get_weather", fn["name"])
+}

backend/internal/pkg/apicompat/anthropic_to_responses.go

@@ -0,0 +1,346 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+)
+
+// AnthropicToResponses converts an Anthropic Messages request directly into
+// a Responses API request. This preserves fields that would be lost in a
+// Chat Completions intermediary round-trip (e.g. thinking, cache_control,
+// structured system prompts).
+func AnthropicToResponses(req *AnthropicRequest) (*ResponsesRequest, error) {
+	input, err := convertAnthropicToResponsesInput(req.System, req.Messages)
+	if err != nil {
+		return nil, err
+	}
+
+	inputJSON, err := json.Marshal(input)
+	if err != nil {
+		return nil, err
+	}
+
+	out := &ResponsesRequest{
+		Model:       req.Model,
+		Input:       inputJSON,
+		Temperature: req.Temperature,
+		TopP:        req.TopP,
+		Stream:      req.Stream,
+		Include:     []string{"reasoning.encrypted_content"},
+	}
+
+	storeFalse := false
+	out.Store = &storeFalse
+
+	if req.MaxTokens > 0 {
+		v := req.MaxTokens
+		if v < minMaxOutputTokens {
+			v = minMaxOutputTokens
+		}
+		out.MaxOutputTokens = &v
+	}
+
+	if len(req.Tools) > 0 {
+		out.Tools = convertAnthropicToolsToResponses(req.Tools)
+	}
+
+	// Convert thinking → reasoning.
+	// generate_summary="auto" causes the upstream to emit reasoning_summary_text
+	// streaming events; the include array only needs reasoning.encrypted_content
+	// (already set above) for content continuity.
+	if req.Thinking != nil {
+		switch req.Thinking.Type {
+		case "enabled":
+			out.Reasoning = &ResponsesReasoning{Effort: "high", Summary: "auto"}
+		case "adaptive":
+			out.Reasoning = &ResponsesReasoning{Effort: "medium", Summary: "auto"}
+		}
+		// "disabled" or unknown → omit reasoning
+	}
+
+	// Convert tool_choice
+	if len(req.ToolChoice) > 0 {
+		tc, err := convertAnthropicToolChoiceToResponses(req.ToolChoice)
+		if err != nil {
+			return nil, fmt.Errorf("convert tool_choice: %w", err)
+		}
+		out.ToolChoice = tc
+	}
+
+	return out, nil
+}
+
+// convertAnthropicToolChoiceToResponses maps Anthropic tool_choice to Responses format.
+//
+//	{"type":"auto"}            → "auto"
+//	{"type":"any"}             → "required"
+//	{"type":"none"}            → "none"
+//	{"type":"tool","name":"X"} → {"type":"function","function":{"name":"X"}}
+func convertAnthropicToolChoiceToResponses(raw json.RawMessage) (json.RawMessage, error) {
+	var tc struct {
+		Type string `json:"type"`
+		Name string `json:"name"`
+	}
+	if err := json.Unmarshal(raw, &tc); err != nil {
+		return nil, err
+	}
+
+	switch tc.Type {
+	case "auto":
+		return json.Marshal("auto")
+	case "any":
+		return json.Marshal("required")
+	case "none":
+		return json.Marshal("none")
+	case "tool":
+		return json.Marshal(map[string]any{
+			"type":     "function",
+			"function": map[string]string{"name": tc.Name},
+		})
+	default:
+		// Pass through unknown types as-is
+		return raw, nil
+	}
+}
+
+// convertAnthropicToResponsesInput builds the Responses API input items array
+// from the Anthropic system field and message list.
+func convertAnthropicToResponsesInput(system json.RawMessage, msgs []AnthropicMessage) ([]ResponsesInputItem, error) {
+	var out []ResponsesInputItem
+
+	// System prompt → system role input item.
+	if len(system) > 0 {
+		sysText, err := parseAnthropicSystemPrompt(system)
+		if err != nil {
+			return nil, err
+		}
+		if sysText != "" {
+			content, _ := json.Marshal(sysText)
+			out = append(out, ResponsesInputItem{
+				Role:    "system",
+				Content: content,
+			})
+		}
+	}
+
+	for _, m := range msgs {
+		items, err := anthropicMsgToResponsesItems(m)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, items...)
+	}
+	return out, nil
+}
+
+// parseAnthropicSystemPrompt handles the Anthropic system field which can be
+// a plain string or an array of text blocks.
+func parseAnthropicSystemPrompt(raw json.RawMessage) (string, error) {
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		return s, nil
+	}
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return "", err
+	}
+	var parts []string
+	for _, b := range blocks {
+		if b.Type == "text" && b.Text != "" {
+			parts = append(parts, b.Text)
+		}
+	}
+	return strings.Join(parts, "\n\n"), nil
+}
+
+// anthropicMsgToResponsesItems converts a single Anthropic message into one
+// or more Responses API input items.
+func anthropicMsgToResponsesItems(m AnthropicMessage) ([]ResponsesInputItem, error) {
+	switch m.Role {
+	case "user":
+		return anthropicUserToResponses(m.Content)
+	case "assistant":
+		return anthropicAssistantToResponses(m.Content)
+	default:
+		return anthropicUserToResponses(m.Content)
+	}
+}
+
+// anthropicUserToResponses handles an Anthropic user message. Content can be a
+// plain string or an array of blocks. tool_result blocks are extracted into
+// function_call_output items.
+func anthropicUserToResponses(raw json.RawMessage) ([]ResponsesInputItem, error) {
+	// Try plain string.
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		content, _ := json.Marshal(s)
+		return []ResponsesInputItem{{Role: "user", Content: content}}, nil
+	}
+
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return nil, err
+	}
+
+	var out []ResponsesInputItem
+
+	// Extract tool_result blocks → function_call_output items.
+	for _, b := range blocks {
+		if b.Type != "tool_result" {
+			continue
+		}
+		text := extractAnthropicToolResultText(b)
+		if text == "" {
+			// OpenAI Responses API requires "output" field; use placeholder for empty results.
+			text = "(empty)"
+		}
+		out = append(out, ResponsesInputItem{
+			Type:   "function_call_output",
+			CallID: toResponsesCallID(b.ToolUseID),
+			Output: text,
+		})
+	}
+
+	// Remaining text blocks → user message.
+	text := extractAnthropicTextFromBlocks(blocks)
+	if text != "" {
+		content, _ := json.Marshal(text)
+		out = append(out, ResponsesInputItem{Role: "user", Content: content})
+	}
+
+	return out, nil
+}
+
+// anthropicAssistantToResponses handles an Anthropic assistant message.
+// Text content → assistant message with output_text parts.
+// tool_use blocks → function_call items.
+// thinking blocks → ignored (OpenAI doesn't accept them as input).
+func anthropicAssistantToResponses(raw json.RawMessage) ([]ResponsesInputItem, error) {
+	// Try plain string.
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		parts := []ResponsesContentPart{{Type: "output_text", Text: s}}
+		partsJSON, err := json.Marshal(parts)
+		if err != nil {
+			return nil, err
+		}
+		return []ResponsesInputItem{{Role: "assistant", Content: partsJSON}}, nil
+	}
+
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return nil, err
+	}
+
+	var items []ResponsesInputItem
+
+	// Text content → assistant message with output_text content parts.
+	text := extractAnthropicTextFromBlocks(blocks)
+	if text != "" {
+		parts := []ResponsesContentPart{{Type: "output_text", Text: text}}
+		partsJSON, err := json.Marshal(parts)
+		if err != nil {
+			return nil, err
+		}
+		items = append(items, ResponsesInputItem{Role: "assistant", Content: partsJSON})
+	}
+
+	// tool_use → function_call items.
+	for _, b := range blocks {
+		if b.Type != "tool_use" {
+			continue
+		}
+		args := "{}"
+		if len(b.Input) > 0 {
+			args = string(b.Input)
+		}
+		fcID := toResponsesCallID(b.ID)
+		items = append(items, ResponsesInputItem{
+			Type:      "function_call",
+			CallID:    fcID,
+			Name:      b.Name,
+			Arguments: args,
+			ID:        fcID,
+		})
+	}
+
+	return items, nil
+}
+
+// toResponsesCallID converts an Anthropic tool ID (toolu_xxx / call_xxx) to a
+// Responses API function_call ID that starts with "fc_".
+func toResponsesCallID(id string) string {
+	if strings.HasPrefix(id, "fc_") {
+		return id
+	}
+	return "fc_" + id
+}
+
+// fromResponsesCallID reverses toResponsesCallID, stripping the "fc_" prefix
+// that was added during request conversion.
+func fromResponsesCallID(id string) string {
+	if after, ok := strings.CutPrefix(id, "fc_"); ok {
+		// Only strip if the remainder doesn't look like it was already "fc_" prefixed.
+		// E.g. "fc_toolu_xxx" → "toolu_xxx", "fc_call_xxx" → "call_xxx"
+		if strings.HasPrefix(after, "toolu_") || strings.HasPrefix(after, "call_") {
+			return after
+		}
+	}
+	return id
+}
+
+// extractAnthropicToolResultText gets the text content from a tool_result block.
+func extractAnthropicToolResultText(b AnthropicContentBlock) string {
+	if len(b.Content) == 0 {
+		return ""
+	}
+	var s string
+	if err := json.Unmarshal(b.Content, &s); err == nil {
+		return s
+	}
+	var inner []AnthropicContentBlock
+	if err := json.Unmarshal(b.Content, &inner); err == nil {
+		var parts []string
+		for _, ib := range inner {
+			if ib.Type == "text" && ib.Text != "" {
+				parts = append(parts, ib.Text)
+			}
+		}
+		return strings.Join(parts, "\n\n")
+	}
+	return ""
+}
+
+// extractAnthropicTextFromBlocks joins all text blocks, ignoring thinking/
+// tool_use/tool_result blocks.
+func extractAnthropicTextFromBlocks(blocks []AnthropicContentBlock) string {
+	var parts []string
+	for _, b := range blocks {
+		if b.Type == "text" && b.Text != "" {
+			parts = append(parts, b.Text)
+		}
+	}
+	return strings.Join(parts, "\n\n")
+}
+
+// convertAnthropicToolsToResponses maps Anthropic tool definitions to
+// Responses API tools. Server-side tools like web_search are mapped to their
+// OpenAI equivalents; regular tools become function tools.
+func convertAnthropicToolsToResponses(tools []AnthropicTool) []ResponsesTool {
+	var out []ResponsesTool
+	for _, t := range tools {
+		// Anthropic server tools like "web_search_20250305" → OpenAI {"type":"web_search"}
+		if strings.HasPrefix(t.Type, "web_search") {
+			out = append(out, ResponsesTool{Type: "web_search"})
+			continue
+		}
+		out = append(out, ResponsesTool{
+			Type:        "function",
+			Name:        t.Name,
+			Description: t.Description,
+			Parameters:  t.InputSchema,
+		})
+	}
+	return out
+}

backend/internal/pkg/apicompat/responses_to_anthropic.go

@@ -0,0 +1,516 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+)
+
+// ---------------------------------------------------------------------------
+// Non-streaming: ResponsesResponse → AnthropicResponse
+// ---------------------------------------------------------------------------
+
+// ResponsesToAnthropic converts a Responses API response directly into an
+// Anthropic Messages response. Reasoning output items are mapped to thinking
+// blocks; function_call items become tool_use blocks.
+func ResponsesToAnthropic(resp *ResponsesResponse, model string) *AnthropicResponse {
+	out := &AnthropicResponse{
+		ID:    resp.ID,
+		Type:  "message",
+		Role:  "assistant",
+		Model: model,
+	}
+
+	var blocks []AnthropicContentBlock
+
+	for _, item := range resp.Output {
+		switch item.Type {
+		case "reasoning":
+			summaryText := ""
+			for _, s := range item.Summary {
+				if s.Type == "summary_text" && s.Text != "" {
+					summaryText += s.Text
+				}
+			}
+			if summaryText != "" {
+				blocks = append(blocks, AnthropicContentBlock{
+					Type:     "thinking",
+					Thinking: summaryText,
+				})
+			}
+		case "message":
+			for _, part := range item.Content {
+				if part.Type == "output_text" && part.Text != "" {
+					blocks = append(blocks, AnthropicContentBlock{
+						Type: "text",
+						Text: part.Text,
+					})
+				}
+			}
+		case "function_call":
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:  "tool_use",
+				ID:    fromResponsesCallID(item.CallID),
+				Name:  item.Name,
+				Input: json.RawMessage(item.Arguments),
+			})
+		case "web_search_call":
+			toolUseID := "srvtoolu_" + item.ID
+			query := ""
+			if item.Action != nil {
+				query = item.Action.Query
+			}
+			inputJSON, _ := json.Marshal(map[string]string{"query": query})
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:  "server_tool_use",
+				ID:    toolUseID,
+				Name:  "web_search",
+				Input: inputJSON,
+			})
+			emptyResults, _ := json.Marshal([]struct{}{})
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:      "web_search_tool_result",
+				ToolUseID: toolUseID,
+				Content:   emptyResults,
+			})
+		}
+	}
+
+	if len(blocks) == 0 {
+		blocks = append(blocks, AnthropicContentBlock{Type: "text", Text: ""})
+	}
+	out.Content = blocks
+
+	out.StopReason = responsesStatusToAnthropicStopReason(resp.Status, resp.IncompleteDetails, blocks)
+
+	if resp.Usage != nil {
+		out.Usage = AnthropicUsage{
+			InputTokens:  resp.Usage.InputTokens,
+			OutputTokens: resp.Usage.OutputTokens,
+		}
+		if resp.Usage.InputTokensDetails != nil {
+			out.Usage.CacheReadInputTokens = resp.Usage.InputTokensDetails.CachedTokens
+		}
+	}
+
+	return out
+}
+
+func responsesStatusToAnthropicStopReason(status string, details *ResponsesIncompleteDetails, blocks []AnthropicContentBlock) string {
+	switch status {
+	case "incomplete":
+		if details != nil && details.Reason == "max_output_tokens" {
+			return "max_tokens"
+		}
+		return "end_turn"
+	case "completed":
+		if len(blocks) > 0 && blocks[len(blocks)-1].Type == "tool_use" {
+			return "tool_use"
+		}
+		return "end_turn"
+	default:
+		return "end_turn"
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Streaming: ResponsesStreamEvent → []AnthropicStreamEvent (stateful converter)
+// ---------------------------------------------------------------------------
+
+// ResponsesEventToAnthropicState tracks state for converting a sequence of
+// Responses SSE events directly into Anthropic SSE events.
+type ResponsesEventToAnthropicState struct {
+	MessageStartSent bool
+	MessageStopSent  bool
+
+	ContentBlockIndex int
+	ContentBlockOpen  bool
+	CurrentBlockType  string // "text" | "thinking" | "tool_use"
+
+	// OutputIndexToBlockIdx maps Responses output_index → Anthropic content block index.
+	OutputIndexToBlockIdx map[int]int
+
+	InputTokens          int
+	OutputTokens         int
+	CacheReadInputTokens int
+
+	ResponseID string
+	Model      string
+	Created    int64
+}
+
+// NewResponsesEventToAnthropicState returns an initialised stream state.
+func NewResponsesEventToAnthropicState() *ResponsesEventToAnthropicState {
+	return &ResponsesEventToAnthropicState{
+		OutputIndexToBlockIdx: make(map[int]int),
+		Created:               time.Now().Unix(),
+	}
+}
+
+// ResponsesEventToAnthropicEvents converts a single Responses SSE event into
+// zero or more Anthropic SSE events, updating state as it goes.
+func ResponsesEventToAnthropicEvents(
+	evt *ResponsesStreamEvent,
+	state *ResponsesEventToAnthropicState,
+) []AnthropicStreamEvent {
+	switch evt.Type {
+	case "response.created":
+		return resToAnthHandleCreated(evt, state)
+	case "response.output_item.added":
+		return resToAnthHandleOutputItemAdded(evt, state)
+	case "response.output_text.delta":
+		return resToAnthHandleTextDelta(evt, state)
+	case "response.output_text.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.function_call_arguments.delta":
+		return resToAnthHandleFuncArgsDelta(evt, state)
+	case "response.function_call_arguments.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.output_item.done":
+		return resToAnthHandleOutputItemDone(evt, state)
+	case "response.reasoning_summary_text.delta":
+		return resToAnthHandleReasoningDelta(evt, state)
+	case "response.reasoning_summary_text.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.completed", "response.incomplete", "response.failed":
+		return resToAnthHandleCompleted(evt, state)
+	default:
+		return nil
+	}
+}
+
+// FinalizeResponsesAnthropicStream emits synthetic termination events if the
+// stream ended without a proper completion event.
+func FinalizeResponsesAnthropicStream(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.MessageStartSent || state.MessageStopSent {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	events = append(events,
+		AnthropicStreamEvent{
+			Type: "message_delta",
+			Delta: &AnthropicDelta{
+				StopReason: "end_turn",
+			},
+			Usage: &AnthropicUsage{
+				InputTokens:          state.InputTokens,
+				OutputTokens:         state.OutputTokens,
+				CacheReadInputTokens: state.CacheReadInputTokens,
+			},
+		},
+		AnthropicStreamEvent{Type: "message_stop"},
+	)
+	state.MessageStopSent = true
+	return events
+}
+
+// ResponsesAnthropicEventToSSE formats an AnthropicStreamEvent as an SSE line pair.
+func ResponsesAnthropicEventToSSE(evt AnthropicStreamEvent) (string, error) {
+	data, err := json.Marshal(evt)
+	if err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("event: %s\ndata: %s\n\n", evt.Type, data), nil
+}
+
+// --- internal handlers ---
+
+func resToAnthHandleCreated(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Response != nil {
+		state.ResponseID = evt.Response.ID
+		// Only use upstream model if no override was set (e.g. originalModel)
+		if state.Model == "" {
+			state.Model = evt.Response.Model
+		}
+	}
+
+	if state.MessageStartSent {
+		return nil
+	}
+	state.MessageStartSent = true
+
+	return []AnthropicStreamEvent{{
+		Type: "message_start",
+		Message: &AnthropicResponse{
+			ID:      state.ResponseID,
+			Type:    "message",
+			Role:    "assistant",
+			Content: []AnthropicContentBlock{},
+			Model:   state.Model,
+			Usage: AnthropicUsage{
+				InputTokens:  0,
+				OutputTokens: 0,
+			},
+		},
+	}}
+}
+
+func resToAnthHandleOutputItemAdded(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Item == nil {
+		return nil
+	}
+
+	switch evt.Item.Type {
+	case "function_call":
+		var events []AnthropicStreamEvent
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.OutputIndexToBlockIdx[evt.OutputIndex] = idx
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "tool_use"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type:  "tool_use",
+				ID:    fromResponsesCallID(evt.Item.CallID),
+				Name:  evt.Item.Name,
+				Input: json.RawMessage("{}"),
+			},
+		})
+		return events
+
+	case "reasoning":
+		var events []AnthropicStreamEvent
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.OutputIndexToBlockIdx[evt.OutputIndex] = idx
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "thinking"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type:     "thinking",
+				Thinking: "",
+			},
+		})
+		return events
+
+	case "message":
+		return nil
+	}
+
+	return nil
+}
+
+func resToAnthHandleTextDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+
+	if !state.ContentBlockOpen || state.CurrentBlockType != "text" {
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "text"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type: "text",
+				Text: "",
+			},
+		})
+	}
+
+	idx := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_delta",
+		Index: &idx,
+		Delta: &AnthropicDelta{
+			Type: "text_delta",
+			Text: evt.Delta,
+		},
+	})
+	return events
+}
+
+func resToAnthHandleFuncArgsDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	blockIdx, ok := state.OutputIndexToBlockIdx[evt.OutputIndex]
+	if !ok {
+		return nil
+	}
+
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_delta",
+		Index: &blockIdx,
+		Delta: &AnthropicDelta{
+			Type:        "input_json_delta",
+			PartialJSON: evt.Delta,
+		},
+	}}
+}
+
+func resToAnthHandleReasoningDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	blockIdx, ok := state.OutputIndexToBlockIdx[evt.OutputIndex]
+	if !ok {
+		return nil
+	}
+
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_delta",
+		Index: &blockIdx,
+		Delta: &AnthropicDelta{
+			Type:     "thinking_delta",
+			Thinking: evt.Delta,
+		},
+	}}
+}
+
+func resToAnthHandleBlockDone(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.ContentBlockOpen {
+		return nil
+	}
+	return closeCurrentBlock(state)
+}
+
+func resToAnthHandleOutputItemDone(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Item == nil {
+		return nil
+	}
+
+	// Handle web_search_call → synthesize server_tool_use + web_search_tool_result blocks.
+	if evt.Item.Type == "web_search_call" && evt.Item.Status == "completed" {
+		return resToAnthHandleWebSearchDone(evt, state)
+	}
+
+	if state.ContentBlockOpen {
+		return closeCurrentBlock(state)
+	}
+	return nil
+}
+
+// resToAnthHandleWebSearchDone converts an OpenAI web_search_call output item
+// into Anthropic server_tool_use + web_search_tool_result content block pairs.
+// This allows Claude Code to count the searches performed.
+func resToAnthHandleWebSearchDone(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	toolUseID := "srvtoolu_" + evt.Item.ID
+	query := ""
+	if evt.Item.Action != nil {
+		query = evt.Item.Action.Query
+	}
+	inputJSON, _ := json.Marshal(map[string]string{"query": query})
+
+	// Emit server_tool_use block (start + stop).
+	idx1 := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_start",
+		Index: &idx1,
+		ContentBlock: &AnthropicContentBlock{
+			Type:  "server_tool_use",
+			ID:    toolUseID,
+			Name:  "web_search",
+			Input: inputJSON,
+		},
+	})
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_stop",
+		Index: &idx1,
+	})
+	state.ContentBlockIndex++
+
+	// Emit web_search_tool_result block (start + stop).
+	// Content is empty because OpenAI does not expose individual search results;
+	// the model consumes them internally and produces text output.
+	emptyResults, _ := json.Marshal([]struct{}{})
+	idx2 := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_start",
+		Index: &idx2,
+		ContentBlock: &AnthropicContentBlock{
+			Type:      "web_search_tool_result",
+			ToolUseID: toolUseID,
+			Content:   emptyResults,
+		},
+	})
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_stop",
+		Index: &idx2,
+	})
+	state.ContentBlockIndex++
+
+	return events
+}
+
+func resToAnthHandleCompleted(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if state.MessageStopSent {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	stopReason := "end_turn"
+	if evt.Response != nil {
+		if evt.Response.Usage != nil {
+			state.InputTokens = evt.Response.Usage.InputTokens
+			state.OutputTokens = evt.Response.Usage.OutputTokens
+			if evt.Response.Usage.InputTokensDetails != nil {
+				state.CacheReadInputTokens = evt.Response.Usage.InputTokensDetails.CachedTokens
+			}
+		}
+		switch evt.Response.Status {
+		case "incomplete":
+			if evt.Response.IncompleteDetails != nil && evt.Response.IncompleteDetails.Reason == "max_output_tokens" {
+				stopReason = "max_tokens"
+			}
+		case "completed":
+			if state.ContentBlockIndex > 0 && state.CurrentBlockType == "tool_use" {
+				stopReason = "tool_use"
+			}
+		}
+	}
+
+	events = append(events,
+		AnthropicStreamEvent{
+			Type: "message_delta",
+			Delta: &AnthropicDelta{
+				StopReason: stopReason,
+			},
+			Usage: &AnthropicUsage{
+				InputTokens:          state.InputTokens,
+				OutputTokens:         state.OutputTokens,
+				CacheReadInputTokens: state.CacheReadInputTokens,
+			},
+		},
+		AnthropicStreamEvent{Type: "message_stop"},
+	)
+	state.MessageStopSent = true
+	return events
+}
+
+func closeCurrentBlock(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.ContentBlockOpen {
+		return nil
+	}
+	idx := state.ContentBlockIndex
+	state.ContentBlockOpen = false
+	state.ContentBlockIndex++
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_stop",
+		Index: &idx,
+	}}
+}

backend/internal/pkg/apicompat/types.go

@@ -0,0 +1,320 @@
+// Package apicompat provides type definitions and conversion utilities for
+// translating between Anthropic Messages and OpenAI Responses API formats.
+// It enables multi-protocol support so that clients using different API
+// formats can be served through a unified gateway.
+package apicompat
+
+import "encoding/json"
+
+// ---------------------------------------------------------------------------
+// Anthropic Messages API types
+// ---------------------------------------------------------------------------
+
+// AnthropicRequest is the request body for POST /v1/messages.
+type AnthropicRequest struct {
+	Model       string             `json:"model"`
+	MaxTokens   int                `json:"max_tokens"`
+	System      json.RawMessage    `json:"system,omitempty"` // string or []AnthropicContentBlock
+	Messages    []AnthropicMessage `json:"messages"`
+	Tools       []AnthropicTool    `json:"tools,omitempty"`
+	Stream      bool               `json:"stream,omitempty"`
+	Temperature *float64           `json:"temperature,omitempty"`
+	TopP        *float64           `json:"top_p,omitempty"`
+	StopSeqs    []string           `json:"stop_sequences,omitempty"`
+	Thinking    *AnthropicThinking `json:"thinking,omitempty"`
+	ToolChoice  json.RawMessage    `json:"tool_choice,omitempty"`
+}
+
+// AnthropicThinking configures extended thinking in the Anthropic API.
+type AnthropicThinking struct {
+	Type         string `json:"type"`                    // "enabled" | "adaptive" | "disabled"
+	BudgetTokens int    `json:"budget_tokens,omitempty"` // max thinking tokens
+}
+
+// AnthropicMessage is a single message in the Anthropic conversation.
+type AnthropicMessage struct {
+	Role    string          `json:"role"` // "user" | "assistant"
+	Content json.RawMessage `json:"content"`
+}
+
+// AnthropicContentBlock is one block inside a message's content array.
+type AnthropicContentBlock struct {
+	Type string `json:"type"`
+
+	// type=text
+	Text string `json:"text,omitempty"`
+
+	// type=thinking
+	Thinking string `json:"thinking,omitempty"`
+
+	// type=tool_use
+	ID    string          `json:"id,omitempty"`
+	Name  string          `json:"name,omitempty"`
+	Input json.RawMessage `json:"input,omitempty"`
+
+	// type=tool_result
+	ToolUseID string          `json:"tool_use_id,omitempty"`
+	Content   json.RawMessage `json:"content,omitempty"` // string or []AnthropicContentBlock
+	IsError   bool            `json:"is_error,omitempty"`
+}
+
+// AnthropicTool describes a tool available to the model.
+type AnthropicTool struct {
+	Type        string          `json:"type,omitempty"`        // e.g. "web_search_20250305" for server tools
+	Name        string          `json:"name"`
+	Description string          `json:"description,omitempty"`
+	InputSchema json.RawMessage `json:"input_schema"` // JSON Schema object
+}
+
+// AnthropicResponse is the non-streaming response from POST /v1/messages.
+type AnthropicResponse struct {
+	ID           string                  `json:"id"`
+	Type         string                  `json:"type"` // "message"
+	Role         string                  `json:"role"` // "assistant"
+	Content      []AnthropicContentBlock `json:"content"`
+	Model        string                  `json:"model"`
+	StopReason   string                  `json:"stop_reason"`
+	StopSequence *string                 `json:"stop_sequence,omitempty"`
+	Usage        AnthropicUsage          `json:"usage"`
+}
+
+// AnthropicUsage holds token counts in Anthropic format.
+type AnthropicUsage struct {
+	InputTokens              int `json:"input_tokens"`
+	OutputTokens             int `json:"output_tokens"`
+	CacheCreationInputTokens int `json:"cache_creation_input_tokens"`
+	CacheReadInputTokens     int `json:"cache_read_input_tokens"`
+}
+
+// ---------------------------------------------------------------------------
+// Anthropic SSE event types
+// ---------------------------------------------------------------------------
+
+// AnthropicStreamEvent is a single SSE event in the Anthropic streaming protocol.
+type AnthropicStreamEvent struct {
+	Type string `json:"type"`
+
+	// message_start
+	Message *AnthropicResponse `json:"message,omitempty"`
+
+	// content_block_start
+	Index        *int                   `json:"index,omitempty"`
+	ContentBlock *AnthropicContentBlock `json:"content_block,omitempty"`
+
+	// content_block_delta
+	Delta *AnthropicDelta `json:"delta,omitempty"`
+
+	// message_delta
+	Usage *AnthropicUsage `json:"usage,omitempty"`
+}
+
+// AnthropicDelta carries incremental content in streaming events.
+type AnthropicDelta struct {
+	Type string `json:"type,omitempty"` // "text_delta" | "input_json_delta" | "thinking_delta" | "signature_delta"
+
+	// text_delta
+	Text string `json:"text,omitempty"`
+
+	// input_json_delta
+	PartialJSON string `json:"partial_json,omitempty"`
+
+	// thinking_delta
+	Thinking string `json:"thinking,omitempty"`
+
+	// signature_delta
+	Signature string `json:"signature,omitempty"`
+
+	// message_delta fields
+	StopReason   string  `json:"stop_reason,omitempty"`
+	StopSequence *string `json:"stop_sequence,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// OpenAI Responses API types
+// ---------------------------------------------------------------------------
+
+// ResponsesRequest is the request body for POST /v1/responses.
+type ResponsesRequest struct {
+	Model           string              `json:"model"`
+	Input           json.RawMessage     `json:"input"` // string or []ResponsesInputItem
+	MaxOutputTokens *int                `json:"max_output_tokens,omitempty"`
+	Temperature     *float64            `json:"temperature,omitempty"`
+	TopP            *float64            `json:"top_p,omitempty"`
+	Stream          bool                `json:"stream,omitempty"`
+	Tools           []ResponsesTool     `json:"tools,omitempty"`
+	Include         []string            `json:"include,omitempty"`
+	Store           *bool               `json:"store,omitempty"`
+	Reasoning       *ResponsesReasoning `json:"reasoning,omitempty"`
+	ToolChoice      json.RawMessage     `json:"tool_choice,omitempty"`
+}
+
+// ResponsesReasoning configures reasoning effort in the Responses API.
+type ResponsesReasoning struct {
+	Effort  string `json:"effort"`            // "low" | "medium" | "high"
+	Summary string `json:"summary,omitempty"` // "auto" | "concise" | "detailed"
+}
+
+// ResponsesInputItem is one item in the Responses API input array.
+// The Type field determines which other fields are populated.
+type ResponsesInputItem struct {
+	// Common
+	Type string `json:"type,omitempty"` // "" for role-based messages
+
+	// Role-based messages (system/user/assistant)
+	Role    string          `json:"role,omitempty"`
+	Content json.RawMessage `json:"content,omitempty"` // string or []ResponsesContentPart
+
+	// type=function_call
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+	ID        string `json:"id,omitempty"`
+
+	// type=function_call_output
+	Output string `json:"output,omitempty"`
+}
+
+// ResponsesContentPart is a typed content part in a Responses message.
+type ResponsesContentPart struct {
+	Type string `json:"type"` // "input_text" | "output_text" | "input_image"
+	Text string `json:"text,omitempty"`
+}
+
+// ResponsesTool describes a tool in the Responses API.
+type ResponsesTool struct {
+	Type        string          `json:"type"` // "function" | "web_search" | "local_shell" etc.
+	Name        string          `json:"name,omitempty"`
+	Description string          `json:"description,omitempty"`
+	Parameters  json.RawMessage `json:"parameters,omitempty"`
+	Strict      *bool           `json:"strict,omitempty"`
+}
+
+// ResponsesResponse is the non-streaming response from POST /v1/responses.
+type ResponsesResponse struct {
+	ID     string            `json:"id"`
+	Object string            `json:"object"` // "response"
+	Model  string            `json:"model"`
+	Status string            `json:"status"` // "completed" | "incomplete" | "failed"
+	Output []ResponsesOutput `json:"output"`
+	Usage  *ResponsesUsage   `json:"usage,omitempty"`
+
+	// incomplete_details is present when status="incomplete"
+	IncompleteDetails *ResponsesIncompleteDetails `json:"incomplete_details,omitempty"`
+
+	// Error is present when status="failed"
+	Error *ResponsesError `json:"error,omitempty"`
+}
+
+// ResponsesError describes an error in a failed response.
+type ResponsesError struct {
+	Code    string `json:"code"`
+	Message string `json:"message"`
+}
+
+// ResponsesIncompleteDetails explains why a response is incomplete.
+type ResponsesIncompleteDetails struct {
+	Reason string `json:"reason"` // "max_output_tokens" | "content_filter"
+}
+
+// ResponsesOutput is one output item in a Responses API response.
+type ResponsesOutput struct {
+	Type string `json:"type"` // "message" | "reasoning" | "function_call" | "web_search_call"
+
+	// type=message
+	ID      string                 `json:"id,omitempty"`
+	Role    string                 `json:"role,omitempty"`
+	Content []ResponsesContentPart `json:"content,omitempty"`
+	Status  string                 `json:"status,omitempty"`
+
+	// type=reasoning
+	EncryptedContent string             `json:"encrypted_content,omitempty"`
+	Summary          []ResponsesSummary `json:"summary,omitempty"`
+
+	// type=function_call
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+
+	// type=web_search_call
+	Action *WebSearchAction `json:"action,omitempty"`
+}
+
+// WebSearchAction describes the search action in a web_search_call output item.
+type WebSearchAction struct {
+	Type  string `json:"type,omitempty"`  // "search"
+	Query string `json:"query,omitempty"` // primary search query
+}
+
+// ResponsesSummary is a summary text block inside a reasoning output.
+type ResponsesSummary struct {
+	Type string `json:"type"` // "summary_text"
+	Text string `json:"text"`
+}
+
+// ResponsesUsage holds token counts in Responses API format.
+type ResponsesUsage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+	TotalTokens  int `json:"total_tokens"`
+
+	// Optional detailed breakdown
+	InputTokensDetails  *ResponsesInputTokensDetails  `json:"input_tokens_details,omitempty"`
+	OutputTokensDetails *ResponsesOutputTokensDetails `json:"output_tokens_details,omitempty"`
+}
+
+// ResponsesInputTokensDetails breaks down input token usage.
+type ResponsesInputTokensDetails struct {
+	CachedTokens int `json:"cached_tokens,omitempty"`
+}
+
+// ResponsesOutputTokensDetails breaks down output token usage.
+type ResponsesOutputTokensDetails struct {
+	ReasoningTokens int `json:"reasoning_tokens,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// Responses SSE event types
+// ---------------------------------------------------------------------------
+
+// ResponsesStreamEvent is a single SSE event in the Responses streaming protocol.
+// The Type field corresponds to the "type" in the JSON payload.
+type ResponsesStreamEvent struct {
+	Type string `json:"type"`
+
+	// response.created / response.completed / response.failed / response.incomplete
+	Response *ResponsesResponse `json:"response,omitempty"`
+
+	// response.output_item.added / response.output_item.done
+	Item *ResponsesOutput `json:"item,omitempty"`
+
+	// response.output_text.delta / response.output_text.done
+	OutputIndex  int    `json:"output_index,omitempty"`
+	ContentIndex int    `json:"content_index,omitempty"`
+	Delta        string `json:"delta,omitempty"`
+	Text         string `json:"text,omitempty"`
+	ItemID       string `json:"item_id,omitempty"`
+
+	// response.function_call_arguments.delta / done
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+
+	// response.reasoning_summary_text.delta / done
+	// Reuses Text/Delta fields above, SummaryIndex identifies which summary part
+	SummaryIndex int `json:"summary_index,omitempty"`
+
+	// error event fields
+	Code  string `json:"code,omitempty"`
+	Param string `json:"param,omitempty"`
+
+	// Sequence number for ordering events
+	SequenceNumber int `json:"sequence_number,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// Shared constants
+// ---------------------------------------------------------------------------
+
+// minMaxOutputTokens is the floor for max_output_tokens in a Responses request.
+// Very small values may cause upstream API errors, so we enforce a minimum.
+const minMaxOutputTokens = 128

backend/internal/pkg/openai/request.go

@@ -58,6 +58,12 @@ func IsCodexOfficialClientOriginator(originator string) bool {
 	return matchCodexClientHeaderPrefixes(v, CodexOfficialClientOriginatorPrefixes)
 }
 
+// IsCodexOfficialClientByHeaders checks whether the request headers indicate an
+// official Codex client family request.
+func IsCodexOfficialClientByHeaders(userAgent, originator string) bool {
+	return IsCodexOfficialClientRequest(userAgent) || IsCodexOfficialClientOriginator(originator)
+}
+
 func normalizeCodexClientHeader(value string) string {
 	return strings.ToLower(strings.TrimSpace(value))
 }

backend/internal/pkg/openai/request_test.go

@@ -85,3 +85,26 @@ func TestIsCodexOfficialClientOriginator(t *testing.T) {
 		})
 	}
 }
+
+func TestIsCodexOfficialClientByHeaders(t *testing.T) {
+	tests := []struct {
+		name       string
+		ua         string
+		originator string
+		want       bool
+	}{
+		{name: "仅 originator 命中 desktop", originator: "Codex Desktop", want: true},
+		{name: "仅 originator 命中 vscode", originator: "codex_vscode", want: true},
+		{name: "仅 ua 命中 desktop", ua: "Codex Desktop/1.2.3", want: true},
+		{name: "ua 与 originator 都未命中", ua: "curl/8.0.1", originator: "my_client", want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := IsCodexOfficialClientByHeaders(tt.ua, tt.originator)
+			if got != tt.want {
+				t.Fatalf("IsCodexOfficialClientByHeaders(%q, %q) = %v, want %v", tt.ua, tt.originator, got, tt.want)
+			}
+		})
+	}
+}

backend/internal/repository/announcement_repo.go

@@ -24,6 +24,7 @@ func (r *announcementRepository) Create(ctx context.Context, a *service.Announce
 		SetTitle(a.Title).
 		SetContent(a.Content).
 		SetStatus(a.Status).
+		SetNotifyMode(a.NotifyMode).
 		SetTargeting(a.Targeting)
 
 	if a.StartsAt != nil {
@@ -64,6 +65,7 @@ func (r *announcementRepository) Update(ctx context.Context, a *service.Announce
 		SetTitle(a.Title).
 		SetContent(a.Content).
 		SetStatus(a.Status).
+		SetNotifyMode(a.NotifyMode).
 		SetTargeting(a.Targeting)
 
 	if a.StartsAt != nil {
@@ -169,17 +171,18 @@ func announcementEntityToService(m *dbent.Announcement) *service.Announcement {
 		return nil
 	}
 	return &service.Announcement{
-		ID:        m.ID,
-		Title:     m.Title,
-		Content:   m.Content,
-		Status:    m.Status,
-		Targeting: m.Targeting,
-		StartsAt:  m.StartsAt,
-		EndsAt:    m.EndsAt,
-		CreatedBy: m.CreatedBy,
-		UpdatedBy: m.UpdatedBy,
-		CreatedAt: m.CreatedAt,
-		UpdatedAt: m.UpdatedAt,
+		ID:         m.ID,
+		Title:      m.Title,
+		Content:    m.Content,
+		Status:     m.Status,
+		NotifyMode: m.NotifyMode,
+		Targeting:  m.Targeting,
+		StartsAt:   m.StartsAt,
+		EndsAt:     m.EndsAt,
+		CreatedBy:  m.CreatedBy,
+		UpdatedBy:  m.UpdatedBy,
+		CreatedAt:  m.CreatedAt,
+		UpdatedAt:  m.UpdatedAt,
 	}
 }
 

backend/internal/repository/api_key_repo.go

@@ -165,6 +165,8 @@ func (r *apiKeyRepository) GetByKeyForAuth(ctx context.Context, key string) (*se
 				group.FieldModelRouting,
 				group.FieldMcpXMLInject,
 				group.FieldSupportedModelScopes,
+			group.FieldAllowMessagesDispatch,
+			group.FieldDefaultMappedModel,
 			)
 		}).
 		Only(ctx)
@@ -470,12 +472,12 @@ func (r *apiKeyRepository) UpdateLastUsed(ctx context.Context, id int64, usedAt
 func (r *apiKeyRepository) IncrementRateLimitUsage(ctx context.Context, id int64, cost float64) error {
 	_, err := r.sql.ExecContext(ctx, `
 		UPDATE api_keys SET
-			usage_5h = usage_5h + $1,
-			usage_1d = usage_1d + $1,
-			usage_7d = usage_7d + $1,
-			window_5h_start = COALESCE(window_5h_start, NOW()),
-			window_1d_start = COALESCE(window_1d_start, NOW()),
-			window_7d_start = COALESCE(window_7d_start, NOW()),
+			usage_5h = CASE WHEN window_5h_start IS NOT NULL AND window_5h_start + INTERVAL '5 hours' <= NOW() THEN $1 ELSE usage_5h + $1 END,
+			usage_1d = CASE WHEN window_1d_start IS NOT NULL AND window_1d_start + INTERVAL '24 hours' <= NOW() THEN $1 ELSE usage_1d + $1 END,
+			usage_7d = CASE WHEN window_7d_start IS NOT NULL AND window_7d_start + INTERVAL '7 days' <= NOW() THEN $1 ELSE usage_7d + $1 END,
+			window_5h_start = CASE WHEN window_5h_start IS NULL OR window_5h_start + INTERVAL '5 hours' <= NOW() THEN NOW() ELSE window_5h_start END,
+			window_1d_start = CASE WHEN window_1d_start IS NULL OR window_1d_start + INTERVAL '24 hours' <= NOW() THEN NOW() ELSE window_1d_start END,
+			window_7d_start = CASE WHEN window_7d_start IS NULL OR window_7d_start + INTERVAL '7 days' <= NOW() THEN NOW() ELSE window_7d_start END,
 			updated_at = NOW()
 		WHERE id = $2 AND deleted_at IS NULL`,
 		cost, id)
@@ -619,6 +621,8 @@ func groupEntityToService(g *dbent.Group) *service.Group {
 		MCPXMLInject:                    g.McpXMLInject,
 		SupportedModelScopes:            g.SupportedModelScopes,
 		SortOrder:                       g.SortOrder,
+		AllowMessagesDispatch:           g.AllowMessagesDispatch,
+		DefaultMappedModel:              g.DefaultMappedModel,
 		CreatedAt:                       g.CreatedAt,
 		UpdatedAt:                       g.UpdatedAt,
 	}

backend/internal/repository/group_repo.go

@@ -59,7 +59,9 @@ func (r *groupRepository) Create(ctx context.Context, groupIn *service.Group) er
 		SetNillableFallbackGroupIDOnInvalidRequest(groupIn.FallbackGroupIDOnInvalidRequest).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
 		SetMcpXMLInject(groupIn.MCPXMLInject).
-		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes)
+		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes).
+		SetAllowMessagesDispatch(groupIn.AllowMessagesDispatch).
+		SetDefaultMappedModel(groupIn.DefaultMappedModel)
 
 	// 设置模型路由配置
 	if groupIn.ModelRouting != nil {
@@ -125,7 +127,9 @@ func (r *groupRepository) Update(ctx context.Context, groupIn *service.Group) er
 		SetClaudeCodeOnly(groupIn.ClaudeCodeOnly).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
 		SetMcpXMLInject(groupIn.MCPXMLInject).
-		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes)
+		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes).
+		SetAllowMessagesDispatch(groupIn.AllowMessagesDispatch).
+		SetDefaultMappedModel(groupIn.DefaultMappedModel)
 
 	// 显式处理可空字段：nil 需要 clear，非 nil 需要 set。
 	if groupIn.DailyLimitUSD != nil {

backend/internal/server/routes/gateway.go

@@ -43,12 +43,33 @@ func RegisterGatewayRoutes(
 	gateway.Use(gin.HandlerFunc(apiKeyAuth))
 	gateway.Use(requireGroupAnthropic)
 	{
-		gateway.POST("/messages", h.Gateway.Messages)
-		gateway.POST("/messages/count_tokens", h.Gateway.CountTokens)
+		// /v1/messages: auto-route based on group platform
+		gateway.POST("/messages", func(c *gin.Context) {
+			if getGroupPlatform(c) == service.PlatformOpenAI {
+				h.OpenAIGateway.Messages(c)
+				return
+			}
+			h.Gateway.Messages(c)
+		})
+		// /v1/messages/count_tokens: OpenAI groups get 404
+		gateway.POST("/messages/count_tokens", func(c *gin.Context) {
+			if getGroupPlatform(c) == service.PlatformOpenAI {
+				c.JSON(http.StatusNotFound, gin.H{
+					"type": "error",
+					"error": gin.H{
+						"type":    "not_found_error",
+						"message": "Token counting is not supported for this platform",
+					},
+				})
+				return
+			}
+			h.Gateway.CountTokens(c)
+		})
 		gateway.GET("/models", h.Gateway.Models)
 		gateway.GET("/usage", h.Gateway.Usage)
 		// OpenAI Responses API
 		gateway.POST("/responses", h.OpenAIGateway.Responses)
+		gateway.POST("/responses/*subpath", h.OpenAIGateway.Responses)
 		gateway.GET("/responses", h.OpenAIGateway.ResponsesWebSocket)
 		// 明确阻止旧协议入口：OpenAI 仅支持 Responses API，避免客户端误解为会自动路由到其它平台。
 		gateway.POST("/chat/completions", func(c *gin.Context) {
@@ -77,6 +98,7 @@ func RegisterGatewayRoutes(
 
 	// OpenAI Responses API（不带v1前缀的别名）
 	r.POST("/responses", bodyLimit, clientRequestID, opsErrorLogger, gin.HandlerFunc(apiKeyAuth), requireGroupAnthropic, h.OpenAIGateway.Responses)
+	r.POST("/responses/*subpath", bodyLimit, clientRequestID, opsErrorLogger, gin.HandlerFunc(apiKeyAuth), requireGroupAnthropic, h.OpenAIGateway.Responses)
 	r.GET("/responses", bodyLimit, clientRequestID, opsErrorLogger, gin.HandlerFunc(apiKeyAuth), requireGroupAnthropic, h.OpenAIGateway.ResponsesWebSocket)
 
 	// Antigravity 模型列表
@@ -132,3 +154,12 @@ func RegisterGatewayRoutes(
 	// Sora 媒体代理（签名 URL，无需 API Key）
 	r.GET("/sora/media-signed/*filepath", h.SoraGateway.MediaProxySigned)
 }
+
+// getGroupPlatform extracts the group platform from the API Key stored in context.
+func getGroupPlatform(c *gin.Context) string {
+	apiKey, ok := middleware.GetAPIKeyFromContext(c)
+	if !ok || apiKey.Group == nil {
+		return ""
+	}
+	return apiKey.Group.Platform
+}

backend/internal/server/routes/gateway_test.go

@@ -0,0 +1,51 @@
+package routes
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/handler"
+	servermiddleware "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+func newGatewayRoutesTestRouter() *gin.Engine {
+	gin.SetMode(gin.TestMode)
+	router := gin.New()
+
+	RegisterGatewayRoutes(
+		router,
+		&handler.Handlers{
+			Gateway:       &handler.GatewayHandler{},
+			OpenAIGateway: &handler.OpenAIGatewayHandler{},
+			SoraGateway:   &handler.SoraGatewayHandler{},
+		},
+		servermiddleware.APIKeyAuthMiddleware(func(c *gin.Context) {
+			c.Next()
+		}),
+		nil,
+		nil,
+		nil,
+		nil,
+		&config.Config{},
+	)
+
+	return router
+}
+
+func TestGatewayRoutesOpenAIResponsesCompactPathIsRegistered(t *testing.T) {
+	router := newGatewayRoutesTestRouter()
+
+	for _, path := range []string{"/v1/responses/compact", "/responses/compact"} {
+		req := httptest.NewRequest(http.MethodPost, path, strings.NewReader(`{"model":"gpt-5"}`))
+		req.Header.Set("Content-Type", "application/json")
+		w := httptest.NewRecorder()
+
+		router.ServeHTTP(w, req)
+		require.NotEqual(t, http.StatusNotFound, w.Code, "path=%s should hit OpenAI responses handler", path)
+	}
+}

backend/internal/service/account_usage_service.go

@@ -1,17 +1,24 @@
 package service
 
 import (
+	"bytes"
 	"context"
+	"encoding/json"
 	"fmt"
 	"log"
+	"math/rand/v2"
+	"net/http"
 	"strings"
 	"sync"
 	"time"
 
+	httppool "github.com/Wei-Shaw/sub2api/internal/pkg/httpclient"
+	openaipkg "github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/timezone"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/usagestats"
 	"golang.org/x/sync/errgroup"
+	"golang.org/x/sync/singleflight"
 )
 
 type UsageLogRepository interface {
@@ -70,8 +77,10 @@ type accountWindowStatsBatchReader interface {
 }
 
 // apiUsageCache 缓存从 Anthropic API 获取的使用率数据（utilization, resets_at）
+// 同时支持缓存错误响应（负缓存），防止 429 等错误导致的重试风暴
 type apiUsageCache struct {
 	response  *ClaudeUsageResponse
+	err       error // 非 nil 表示缓存的错误（负缓存）
 	timestamp time.Time
 }
 
@@ -88,15 +97,21 @@ type antigravityUsageCache struct {
 }
 
 const (
-	apiCacheTTL         = 3 * time.Minute
-	windowStatsCacheTTL = 1 * time.Minute
+	apiCacheTTL             = 3 * time.Minute
+	apiErrorCacheTTL        = 1 * time.Minute        // 负缓存 TTL：429 等错误缓存 1 分钟
+	apiQueryMaxJitter       = 800 * time.Millisecond // 用量查询最大随机延迟
+	windowStatsCacheTTL     = 1 * time.Minute
+	openAIProbeCacheTTL     = 10 * time.Minute
+	openAICodexProbeVersion = "0.104.0"
 )
 
 // UsageCache 封装账户使用量相关的缓存
 type UsageCache struct {
-	apiCache         sync.Map // accountID -> *apiUsageCache
-	windowStatsCache sync.Map // accountID -> *windowStatsCache
-	antigravityCache sync.Map // accountID -> *antigravityUsageCache
+	apiCache         sync.Map           // accountID -> *apiUsageCache
+	windowStatsCache sync.Map           // accountID -> *windowStatsCache
+	antigravityCache sync.Map           // accountID -> *antigravityUsageCache
+	apiFlight        singleflight.Group // 防止同一账号的并发请求击穿缓存
+	openAIProbeCache sync.Map           // accountID -> time.Time
 }
 
 // NewUsageCache 创建 UsageCache 实例
@@ -224,6 +239,14 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64) (*U
 		return nil, fmt.Errorf("get account failed: %w", err)
 	}
 
+	if account.Platform == PlatformOpenAI && account.Type == AccountTypeOAuth {
+		usage, err := s.getOpenAIUsage(ctx, account)
+		if err == nil {
+			s.tryClearRecoverableAccountError(ctx, account)
+		}
+		return usage, err
+	}
+
 	if account.Platform == PlatformGemini {
 		usage, err := s.getGeminiUsage(ctx, account)
 		if err == nil {
@@ -245,24 +268,65 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64) (*U
 	if account.CanGetUsage() {
 		var apiResp *ClaudeUsageResponse
 
-		// 1. 检查 API 缓存（10 分钟）
+		// 1. 检查缓存（成功响应 3 分钟 / 错误响应 1 分钟）
 		if cached, ok := s.cache.apiCache.Load(accountID); ok {
-			if cache, ok := cached.(*apiUsageCache); ok && time.Since(cache.timestamp) < apiCacheTTL {
-				apiResp = cache.response
+			if cache, ok := cached.(*apiUsageCache); ok {
+				age := time.Since(cache.timestamp)
+				if cache.err != nil && age < apiErrorCacheTTL {
+					// 负缓存命中：返回缓存的错误，避免重试风暴
+					return nil, cache.err
+				}
+				if cache.response != nil && age < apiCacheTTL {
+					apiResp = cache.response
+				}
 			}
 		}
 
-		// 2. 如果没有缓存，从 API 获取
+		// 2. 如果没有有效缓存，通过 singleflight 从 API 获取（防止并发击穿）
 		if apiResp == nil {
-			apiResp, err = s.fetchOAuthUsageRaw(ctx, account)
-			if err != nil {
-				return nil, err
+			// 随机延迟：打散多账号并发请求，避免同一时刻大量相同 TLS 指纹请求
+			// 触发上游反滥用检测。延迟范围 0~800ms，仅在缓存未命中时生效。
+			jitter := time.Duration(rand.Int64N(int64(apiQueryMaxJitter)))
+			select {
+			case <-time.After(jitter):
+			case <-ctx.Done():
+				return nil, ctx.Err()
 			}
-			// 缓存 API 响应
-			s.cache.apiCache.Store(accountID, &apiUsageCache{
-				response:  apiResp,
-				timestamp: time.Now(),
+
+			flightKey := fmt.Sprintf("usage:%d", accountID)
+			result, flightErr, _ := s.cache.apiFlight.Do(flightKey, func() (any, error) {
+				// 再次检查缓存（可能在等待 singleflight 期间被其他请求填充）
+				if cached, ok := s.cache.apiCache.Load(accountID); ok {
+					if cache, ok := cached.(*apiUsageCache); ok {
+						age := time.Since(cache.timestamp)
+						if cache.err != nil && age < apiErrorCacheTTL {
+							return nil, cache.err
+						}
+						if cache.response != nil && age < apiCacheTTL {
+							return cache.response, nil
+						}
+					}
+				}
+				resp, fetchErr := s.fetchOAuthUsageRaw(ctx, account)
+				if fetchErr != nil {
+					// 负缓存：缓存错误响应，防止后续请求重复触发 429
+					s.cache.apiCache.Store(accountID, &apiUsageCache{
+						err:       fetchErr,
+						timestamp: time.Now(),
+					})
+					return nil, fetchErr
+				}
+				// 缓存成功响应
+				s.cache.apiCache.Store(accountID, &apiUsageCache{
+					response:  resp,
+					timestamp: time.Now(),
+				})
+				return resp, nil
 			})
+			if flightErr != nil {
+				return nil, flightErr
+			}
+			apiResp, _ = result.(*ClaudeUsageResponse)
 		}
 
 		// 3. 构建 UsageInfo（每次都重新计算 RemainingSeconds）
@@ -288,6 +352,161 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64) (*U
 	return nil, fmt.Errorf("account type %s does not support usage query", account.Type)
 }
 
+func (s *AccountUsageService) getOpenAIUsage(ctx context.Context, account *Account) (*UsageInfo, error) {
+	now := time.Now()
+	usage := &UsageInfo{UpdatedAt: &now}
+
+	if account == nil {
+		return usage, nil
+	}
+
+	if progress := buildCodexUsageProgressFromExtra(account.Extra, "5h", now); progress != nil {
+		usage.FiveHour = progress
+	}
+	if progress := buildCodexUsageProgressFromExtra(account.Extra, "7d", now); progress != nil {
+		usage.SevenDay = progress
+	}
+
+	if (usage.FiveHour == nil || usage.SevenDay == nil) && s.shouldProbeOpenAICodexSnapshot(account.ID, now) {
+		if updates, err := s.probeOpenAICodexSnapshot(ctx, account); err == nil && len(updates) > 0 {
+			mergeAccountExtra(account, updates)
+			if usage.UpdatedAt == nil {
+				usage.UpdatedAt = &now
+			}
+			if progress := buildCodexUsageProgressFromExtra(account.Extra, "5h", now); progress != nil {
+				usage.FiveHour = progress
+			}
+			if progress := buildCodexUsageProgressFromExtra(account.Extra, "7d", now); progress != nil {
+				usage.SevenDay = progress
+			}
+		}
+	}
+
+	if s.usageLogRepo == nil {
+		return usage, nil
+	}
+
+	if stats, err := s.usageLogRepo.GetAccountWindowStats(ctx, account.ID, now.Add(-5*time.Hour)); err == nil {
+		windowStats := windowStatsFromAccountStats(stats)
+		if hasMeaningfulWindowStats(windowStats) {
+			if usage.FiveHour == nil {
+				usage.FiveHour = &UsageProgress{Utilization: 0}
+			}
+			usage.FiveHour.WindowStats = windowStats
+		}
+	}
+
+	if stats, err := s.usageLogRepo.GetAccountWindowStats(ctx, account.ID, now.Add(-7*24*time.Hour)); err == nil {
+		windowStats := windowStatsFromAccountStats(stats)
+		if hasMeaningfulWindowStats(windowStats) {
+			if usage.SevenDay == nil {
+				usage.SevenDay = &UsageProgress{Utilization: 0}
+			}
+			usage.SevenDay.WindowStats = windowStats
+		}
+	}
+
+	return usage, nil
+}
+
+func (s *AccountUsageService) shouldProbeOpenAICodexSnapshot(accountID int64, now time.Time) bool {
+	if s == nil || s.cache == nil || accountID <= 0 {
+		return true
+	}
+	if cached, ok := s.cache.openAIProbeCache.Load(accountID); ok {
+		if ts, ok := cached.(time.Time); ok && now.Sub(ts) < openAIProbeCacheTTL {
+			return false
+		}
+	}
+	s.cache.openAIProbeCache.Store(accountID, now)
+	return true
+}
+
+func (s *AccountUsageService) probeOpenAICodexSnapshot(ctx context.Context, account *Account) (map[string]any, error) {
+	if account == nil || !account.IsOAuth() {
+		return nil, nil
+	}
+	accessToken := account.GetOpenAIAccessToken()
+	if accessToken == "" {
+		return nil, fmt.Errorf("no access token available")
+	}
+	modelID := openaipkg.DefaultTestModel
+	payload := createOpenAITestPayload(modelID, true)
+	payloadBytes, err := json.Marshal(payload)
+	if err != nil {
+		return nil, fmt.Errorf("marshal openai probe payload: %w", err)
+	}
+
+	reqCtx, cancel := context.WithTimeout(ctx, 15*time.Second)
+	defer cancel()
+	req, err := http.NewRequestWithContext(reqCtx, http.MethodPost, chatgptCodexURL, bytes.NewReader(payloadBytes))
+	if err != nil {
+		return nil, fmt.Errorf("create openai probe request: %w", err)
+	}
+	req.Host = "chatgpt.com"
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer "+accessToken)
+	req.Header.Set("Accept", "text/event-stream")
+	req.Header.Set("OpenAI-Beta", "responses=experimental")
+	req.Header.Set("Originator", "codex_cli_rs")
+	req.Header.Set("Version", openAICodexProbeVersion)
+	req.Header.Set("User-Agent", codexCLIUserAgent)
+	if s.identityCache != nil {
+		if fp, fpErr := s.identityCache.GetFingerprint(reqCtx, account.ID); fpErr == nil && fp != nil && strings.TrimSpace(fp.UserAgent) != "" {
+			req.Header.Set("User-Agent", strings.TrimSpace(fp.UserAgent))
+		}
+	}
+	if chatgptAccountID := account.GetChatGPTAccountID(); chatgptAccountID != "" {
+		req.Header.Set("chatgpt-account-id", chatgptAccountID)
+	}
+
+	proxyURL := ""
+	if account.ProxyID != nil && account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	client, err := httppool.GetClient(httppool.Options{
+		ProxyURL:              proxyURL,
+		Timeout:               15 * time.Second,
+		ResponseHeaderTimeout: 10 * time.Second,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("build openai probe client: %w", err)
+	}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("openai codex probe request failed: %w", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		return nil, fmt.Errorf("openai codex probe returned status %d", resp.StatusCode)
+	}
+	if snapshot := ParseCodexRateLimitHeaders(resp.Header); snapshot != nil {
+		updates := buildCodexUsageExtraUpdates(snapshot, time.Now())
+		if len(updates) > 0 {
+			go func(accountID int64, updates map[string]any) {
+				updateCtx, updateCancel := context.WithTimeout(context.Background(), 5*time.Second)
+				defer updateCancel()
+				_ = s.accountRepo.UpdateExtra(updateCtx, accountID, updates)
+			}(account.ID, updates)
+			return updates, nil
+		}
+	}
+	return nil, nil
+}
+
+func mergeAccountExtra(account *Account, updates map[string]any) {
+	if account == nil || len(updates) == 0 {
+		return
+	}
+	if account.Extra == nil {
+		account.Extra = make(map[string]any, len(updates))
+	}
+	for k, v := range updates {
+		account.Extra[k] = v
+	}
+}
+
 func (s *AccountUsageService) getGeminiUsage(ctx context.Context, account *Account) (*UsageInfo, error) {
 	now := time.Now()
 	usage := &UsageInfo{
@@ -519,6 +738,72 @@ func windowStatsFromAccountStats(stats *usagestats.AccountStats) *WindowStats {
 	}
 }
 
+func hasMeaningfulWindowStats(stats *WindowStats) bool {
+	if stats == nil {
+		return false
+	}
+	return stats.Requests > 0 || stats.Tokens > 0 || stats.Cost > 0 || stats.StandardCost > 0 || stats.UserCost > 0
+}
+
+func buildCodexUsageProgressFromExtra(extra map[string]any, window string, now time.Time) *UsageProgress {
+	if len(extra) == 0 {
+		return nil
+	}
+
+	var (
+		usedPercentKey string
+		resetAfterKey  string
+		resetAtKey     string
+	)
+
+	switch window {
+	case "5h":
+		usedPercentKey = "codex_5h_used_percent"
+		resetAfterKey = "codex_5h_reset_after_seconds"
+		resetAtKey = "codex_5h_reset_at"
+	case "7d":
+		usedPercentKey = "codex_7d_used_percent"
+		resetAfterKey = "codex_7d_reset_after_seconds"
+		resetAtKey = "codex_7d_reset_at"
+	default:
+		return nil
+	}
+
+	usedRaw, ok := extra[usedPercentKey]
+	if !ok {
+		return nil
+	}
+
+	progress := &UsageProgress{Utilization: parseExtraFloat64(usedRaw)}
+	if resetAtRaw, ok := extra[resetAtKey]; ok {
+		if resetAt, err := parseTime(fmt.Sprint(resetAtRaw)); err == nil {
+			progress.ResetsAt = &resetAt
+			progress.RemainingSeconds = int(time.Until(resetAt).Seconds())
+			if progress.RemainingSeconds < 0 {
+				progress.RemainingSeconds = 0
+			}
+		}
+	}
+	if progress.ResetsAt == nil {
+		if resetAfterSeconds := parseExtraInt(extra[resetAfterKey]); resetAfterSeconds > 0 {
+			base := now
+			if updatedAtRaw, ok := extra["codex_usage_updated_at"]; ok {
+				if updatedAt, err := parseTime(fmt.Sprint(updatedAtRaw)); err == nil {
+					base = updatedAt
+				}
+			}
+			resetAt := base.Add(time.Duration(resetAfterSeconds) * time.Second)
+			progress.ResetsAt = &resetAt
+			progress.RemainingSeconds = int(time.Until(resetAt).Seconds())
+			if progress.RemainingSeconds < 0 {
+				progress.RemainingSeconds = 0
+			}
+		}
+	}
+
+	return progress
+}
+
 func (s *AccountUsageService) GetAccountUsageStats(ctx context.Context, accountID int64, startTime, endTime time.Time) (*usagestats.AccountUsageStatsResponse, error) {
 	stats, err := s.usageLogRepo.GetAccountUsageStats(ctx, accountID, startTime, endTime)
 	if err != nil {
@@ -666,15 +951,30 @@ func (s *AccountUsageService) estimateSetupTokenUsage(account *Account) *UsageIn
 			remaining = 0
 		}
 
-		// 根据状态估算使用率 (百分比形式，100 = 100%)
+		// 优先使用响应头中存储的真实 utilization 值（0-1 小数，转为 0-100 百分比）
 		var utilization float64
-		switch account.SessionWindowStatus {
-		case "rejected":
-			utilization = 100.0
-		case "allowed_warning":
-			utilization = 80.0
-		default:
-			utilization = 0.0
+		var found bool
+		if stored, ok := account.Extra["session_window_utilization"]; ok {
+			switch v := stored.(type) {
+			case float64:
+				utilization = v * 100
+				found = true
+			case json.Number:
+				if f, err := v.Float64(); err == nil {
+					utilization = f * 100
+					found = true
+				}
+			}
+		}
+
+		// 如果没有存储的 utilization，回退到状态估算
+		if !found {
+			switch account.SessionWindowStatus {
+			case "rejected":
+				utilization = 100.0
+			case "allowed_warning":
+				utilization = 80.0
+			}
 		}
 
 		info.FiveHour = &UsageProgress{

backend/internal/service/admin_service.go

@@ -145,6 +145,9 @@ type CreateGroupInput struct {
 	SupportedModelScopes []string
 	// Sora 存储配额
 	SoraStorageQuotaBytes int64
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch bool
+	DefaultMappedModel    string
 	// 从指定分组复制账号（创建分组后在同一事务内绑定）
 	CopyAccountsFromGroupIDs []int64
 }
@@ -181,6 +184,9 @@ type UpdateGroupInput struct {
 	SupportedModelScopes *[]string
 	// Sora 存储配额
 	SoraStorageQuotaBytes *int64
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch *bool
+	DefaultMappedModel    *string
 	// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
 	CopyAccountsFromGroupIDs []int64
 }
@@ -909,6 +915,8 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 		MCPXMLInject:                    mcpXMLInject,
 		SupportedModelScopes:            input.SupportedModelScopes,
 		SoraStorageQuotaBytes:           input.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           input.AllowMessagesDispatch,
+		DefaultMappedModel:              input.DefaultMappedModel,
 	}
 	if err := s.groupRepo.Create(ctx, group); err != nil {
 		return nil, err
@@ -1122,6 +1130,14 @@ func (s *adminServiceImpl) UpdateGroup(ctx context.Context, id int64, input *Upd
 		group.SupportedModelScopes = *input.SupportedModelScopes
 	}
 
+	// OpenAI Messages 调度配置
+	if input.AllowMessagesDispatch != nil {
+		group.AllowMessagesDispatch = *input.AllowMessagesDispatch
+	}
+	if input.DefaultMappedModel != nil {
+		group.DefaultMappedModel = *input.DefaultMappedModel
+	}
+
 	if err := s.groupRepo.Update(ctx, group); err != nil {
 		return nil, err
 	}

backend/internal/service/announcement.go

@@ -14,6 +14,11 @@ const (
 	AnnouncementStatusArchived = domain.AnnouncementStatusArchived
 )
 
+const (
+	AnnouncementNotifyModeSilent = domain.AnnouncementNotifyModeSilent
+	AnnouncementNotifyModePopup  = domain.AnnouncementNotifyModePopup
+)
+
 const (
 	AnnouncementConditionTypeSubscription = domain.AnnouncementConditionTypeSubscription
 	AnnouncementConditionTypeBalance      = domain.AnnouncementConditionTypeBalance

backend/internal/service/announcement_service.go

@@ -33,23 +33,25 @@ func NewAnnouncementService(
 }
 
 type CreateAnnouncementInput struct {
-	Title     string
-	Content   string
-	Status    string
-	Targeting AnnouncementTargeting
-	StartsAt  *time.Time
-	EndsAt    *time.Time
-	ActorID   *int64 // 管理员用户ID
+	Title      string
+	Content    string
+	Status     string
+	NotifyMode string
+	Targeting  AnnouncementTargeting
+	StartsAt   *time.Time
+	EndsAt     *time.Time
+	ActorID    *int64 // 管理员用户ID
 }
 
 type UpdateAnnouncementInput struct {
-	Title     *string
-	Content   *string
-	Status    *string
-	Targeting *AnnouncementTargeting
-	StartsAt  **time.Time
-	EndsAt    **time.Time
-	ActorID   *int64 // 管理员用户ID
+	Title      *string
+	Content    *string
+	Status     *string
+	NotifyMode *string
+	Targeting  *AnnouncementTargeting
+	StartsAt   **time.Time
+	EndsAt     **time.Time
+	ActorID    *int64 // 管理员用户ID
 }
 
 type UserAnnouncement struct {
@@ -93,6 +95,14 @@ func (s *AnnouncementService) Create(ctx context.Context, input *CreateAnnouncem
 		return nil, err
 	}
 
+	notifyMode := strings.TrimSpace(input.NotifyMode)
+	if notifyMode == "" {
+		notifyMode = AnnouncementNotifyModeSilent
+	}
+	if !isValidAnnouncementNotifyMode(notifyMode) {
+		return nil, fmt.Errorf("create announcement: invalid notify_mode")
+	}
+
 	if input.StartsAt != nil && input.EndsAt != nil {
 		if !input.StartsAt.Before(*input.EndsAt) {
 			return nil, fmt.Errorf("create announcement: starts_at must be before ends_at")
@@ -100,12 +110,13 @@ func (s *AnnouncementService) Create(ctx context.Context, input *CreateAnnouncem
 	}
 
 	a := &Announcement{
-		Title:     title,
-		Content:   content,
-		Status:    status,
-		Targeting: targeting,
-		StartsAt:  input.StartsAt,
-		EndsAt:    input.EndsAt,
+		Title:      title,
+		Content:    content,
+		Status:     status,
+		NotifyMode: notifyMode,
+		Targeting:  targeting,
+		StartsAt:   input.StartsAt,
+		EndsAt:     input.EndsAt,
 	}
 	if input.ActorID != nil && *input.ActorID > 0 {
 		a.CreatedBy = input.ActorID
@@ -150,6 +161,14 @@ func (s *AnnouncementService) Update(ctx context.Context, id int64, input *Updat
 		a.Status = status
 	}
 
+	if input.NotifyMode != nil {
+		notifyMode := strings.TrimSpace(*input.NotifyMode)
+		if !isValidAnnouncementNotifyMode(notifyMode) {
+			return nil, fmt.Errorf("update announcement: invalid notify_mode")
+		}
+		a.NotifyMode = notifyMode
+	}
+
 	if input.Targeting != nil {
 		targeting, err := domain.AnnouncementTargeting(*input.Targeting).NormalizeAndValidate()
 		if err != nil {
@@ -376,3 +395,12 @@ func isValidAnnouncementStatus(status string) bool {
 		return false
 	}
 }
+
+func isValidAnnouncementNotifyMode(mode string) bool {
+	switch mode {
+	case AnnouncementNotifyModeSilent, AnnouncementNotifyModePopup:
+		return true
+	default:
+		return false
+	}
+}

backend/internal/service/antigravity_gateway_service.go

@@ -3696,6 +3696,15 @@ func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context
 				finalEvents, agUsage := processor.Finish()
 				if len(finalEvents) > 0 {
 					cw.Write(finalEvents)
+				} else if !processor.MessageStartSent() && !cw.Disconnected() {
+					// 整个流未收到任何可解析的上游数据（全部 SSE 行均无法被 JSON 解析），
+					// 触发 failover 在同账号重试，避免向客户端发出缺少 message_start 的残缺流
+					logger.LegacyPrintf("service.antigravity_gateway", "[antigravity-Claude-Stream] empty stream response (no valid events parsed), triggering failover")
+					return nil, &UpstreamFailoverError{
+						StatusCode:             http.StatusBadGateway,
+						ResponseBody:           []byte(`{"error":"empty stream response from upstream"}`),
+						RetryableOnSameAccount: true,
+					}
 				}
 				return &antigravityStreamResult{usage: convertUsage(agUsage), firstTokenMs: firstTokenMs, clientDisconnect: cw.Disconnected()}, nil
 			}

backend/internal/service/antigravity_gateway_service_test.go

@@ -998,6 +998,46 @@ func TestHandleClaudeStreamingResponse_ClientDisconnect(t *testing.T) {
 	require.True(t, result.clientDisconnect)
 }
 
+// TestHandleClaudeStreamingResponse_EmptyStream
+// 验证：上游只返回无法解析的 SSE 行时，触发 UpstreamFailoverError 而不是向客户端发出残缺流
+func TestHandleClaudeStreamingResponse_EmptyStream(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newAntigravityTestService(&config.Config{
+		Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize},
+	})
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	pr, pw := io.Pipe()
+	resp := &http.Response{StatusCode: http.StatusOK, Body: pr, Header: http.Header{}}
+
+	go func() {
+		defer func() { _ = pw.Close() }()
+		// 所有行均为无法 JSON 解析的内容，ProcessLine 全部返回 nil
+		fmt.Fprintln(pw, "data: not-valid-json")
+		fmt.Fprintln(pw, "")
+		fmt.Fprintln(pw, "data: also-invalid")
+		fmt.Fprintln(pw, "")
+	}()
+
+	_, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5")
+	_ = pr.Close()
+
+	// 应当返回 UpstreamFailoverError 而非 nil，以便上层触发 failover
+	require.Error(t, err)
+	var failoverErr *UpstreamFailoverError
+	require.ErrorAs(t, err, &failoverErr)
+	require.True(t, failoverErr.RetryableOnSameAccount)
+
+	// 客户端不应收到任何 SSE 事件（既无 message_start 也无 message_stop）
+	body := rec.Body.String()
+	require.NotContains(t, body, "event: message_start")
+	require.NotContains(t, body, "event: message_stop")
+	require.NotContains(t, body, "event: message_delta")
+}
+
 // TestHandleClaudeStreamingResponse_ContextCanceled
 // 验证：context 取消时不注入错误事件
 func TestHandleClaudeStreamingResponse_ContextCanceled(t *testing.T) {

backend/internal/service/api_key.go

@@ -14,6 +14,18 @@ const (
 	StatusAPIKeyExpired        = "expired"
 )
 
+// Rate limit window durations
+const (
+	RateLimitWindow5h = 5 * time.Hour
+	RateLimitWindow1d = 24 * time.Hour
+	RateLimitWindow7d = 7 * 24 * time.Hour
+)
+
+// IsWindowExpired returns true if the window starting at windowStart has exceeded the given duration.
+func IsWindowExpired(windowStart *time.Time, duration time.Duration) bool {
+	return windowStart != nil && time.Since(*windowStart) >= duration
+}
+
 type APIKey struct {
 	ID          int64
 	UserID      int64
@@ -98,6 +110,30 @@ func (k *APIKey) GetDaysUntilExpiry() int {
 	return int(duration.Hours() / 24)
 }
 
+// EffectiveUsage5h returns the 5h window usage, or 0 if the window has expired.
+func (k *APIKey) EffectiveUsage5h() float64 {
+	if IsWindowExpired(k.Window5hStart, RateLimitWindow5h) {
+		return 0
+	}
+	return k.Usage5h
+}
+
+// EffectiveUsage1d returns the 1d window usage, or 0 if the window has expired.
+func (k *APIKey) EffectiveUsage1d() float64 {
+	if IsWindowExpired(k.Window1dStart, RateLimitWindow1d) {
+		return 0
+	}
+	return k.Usage1d
+}
+
+// EffectiveUsage7d returns the 7d window usage, or 0 if the window has expired.
+func (k *APIKey) EffectiveUsage7d() float64 {
+	if IsWindowExpired(k.Window7dStart, RateLimitWindow7d) {
+		return 0
+	}
+	return k.Usage7d
+}
+
 // APIKeyListFilters holds optional filtering parameters for listing API keys.
 type APIKeyListFilters struct {
 	Search  string

backend/internal/service/api_key_auth_cache.go

@@ -65,6 +65,10 @@ type APIKeyAuthGroupSnapshot struct {
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`
+
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch bool   `json:"allow_messages_dispatch"`
+	DefaultMappedModel    string `json:"default_mapped_model,omitempty"`
 }
 
 // APIKeyAuthCacheEntry 缓存条目，支持负缓存

backend/internal/service/api_key_auth_cache_impl.go

@@ -245,6 +245,8 @@ func (s *APIKeyService) snapshotFromAPIKey(apiKey *APIKey) *APIKeyAuthSnapshot {
 			ModelRoutingEnabled:             apiKey.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    apiKey.Group.MCPXMLInject,
 			SupportedModelScopes:            apiKey.Group.SupportedModelScopes,
+			AllowMessagesDispatch:           apiKey.Group.AllowMessagesDispatch,
+			DefaultMappedModel:              apiKey.Group.DefaultMappedModel,
 		}
 	}
 	return snapshot
@@ -302,6 +304,8 @@ func (s *APIKeyService) snapshotToAPIKey(key string, snapshot *APIKeyAuthSnapsho
 			ModelRoutingEnabled:             snapshot.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    snapshot.Group.MCPXMLInject,
 			SupportedModelScopes:            snapshot.Group.SupportedModelScopes,
+			AllowMessagesDispatch:           snapshot.Group.AllowMessagesDispatch,
+			DefaultMappedModel:              snapshot.Group.DefaultMappedModel,
 		}
 	}
 	s.compileAPIKeyIPRules(apiKey)

backend/internal/service/api_key_rate_limit_test.go

@@ -0,0 +1,245 @@
+package service
+
+import (
+	"testing"
+	"time"
+)
+
+func TestIsWindowExpired(t *testing.T) {
+	now := time.Now()
+
+	tests := []struct {
+		name     string
+		start    *time.Time
+		duration time.Duration
+		want     bool
+	}{
+		{
+			name:     "nil window start",
+			start:    nil,
+			duration: RateLimitWindow5h,
+			want:     false,
+		},
+		{
+			name:     "active window (started 1h ago, 5h window)",
+			start:    rateLimitTimePtr(now.Add(-1 * time.Hour)),
+			duration: RateLimitWindow5h,
+			want:     false,
+		},
+		{
+			name:     "expired window (started 6h ago, 5h window)",
+			start:    rateLimitTimePtr(now.Add(-6 * time.Hour)),
+			duration: RateLimitWindow5h,
+			want:     true,
+		},
+		{
+			name:     "exactly at boundary (started 5h ago, 5h window)",
+			start:    rateLimitTimePtr(now.Add(-5 * time.Hour)),
+			duration: RateLimitWindow5h,
+			want:     true,
+		},
+		{
+			name:     "active 1d window (started 12h ago)",
+			start:    rateLimitTimePtr(now.Add(-12 * time.Hour)),
+			duration: RateLimitWindow1d,
+			want:     false,
+		},
+		{
+			name:     "expired 1d window (started 25h ago)",
+			start:    rateLimitTimePtr(now.Add(-25 * time.Hour)),
+			duration: RateLimitWindow1d,
+			want:     true,
+		},
+		{
+			name:     "active 7d window (started 3d ago)",
+			start:    rateLimitTimePtr(now.Add(-3 * 24 * time.Hour)),
+			duration: RateLimitWindow7d,
+			want:     false,
+		},
+		{
+			name:     "expired 7d window (started 8d ago)",
+			start:    rateLimitTimePtr(now.Add(-8 * 24 * time.Hour)),
+			duration: RateLimitWindow7d,
+			want:     true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := IsWindowExpired(tt.start, tt.duration)
+			if got != tt.want {
+				t.Errorf("IsWindowExpired() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
+
+func TestAPIKey_EffectiveUsage(t *testing.T) {
+	now := time.Now()
+
+	tests := []struct {
+		name      string
+		key       APIKey
+		want5h    float64
+		want1d    float64
+		want7d    float64
+	}{
+		{
+			name: "all windows active",
+			key: APIKey{
+				Usage5h:       5.0,
+				Usage1d:       10.0,
+				Usage7d:       50.0,
+				Window5hStart: rateLimitTimePtr(now.Add(-1 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-12 * time.Hour)),
+				Window7dStart: rateLimitTimePtr(now.Add(-3 * 24 * time.Hour)),
+			},
+			want5h: 5.0,
+			want1d: 10.0,
+			want7d: 50.0,
+		},
+		{
+			name: "all windows expired",
+			key: APIKey{
+				Usage5h:       5.0,
+				Usage1d:       10.0,
+				Usage7d:       50.0,
+				Window5hStart: rateLimitTimePtr(now.Add(-6 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-25 * time.Hour)),
+				Window7dStart: rateLimitTimePtr(now.Add(-8 * 24 * time.Hour)),
+			},
+			want5h: 0,
+			want1d: 0,
+			want7d: 0,
+		},
+		{
+			name: "nil window starts return raw usage",
+			key: APIKey{
+				Usage5h:       5.0,
+				Usage1d:       10.0,
+				Usage7d:       50.0,
+				Window5hStart: nil,
+				Window1dStart: nil,
+				Window7dStart: nil,
+			},
+			want5h: 5.0,
+			want1d: 10.0,
+			want7d: 50.0,
+		},
+		{
+			name: "mixed: 5h expired, 1d active, 7d nil",
+			key: APIKey{
+				Usage5h:       5.0,
+				Usage1d:       10.0,
+				Usage7d:       50.0,
+				Window5hStart: rateLimitTimePtr(now.Add(-6 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-12 * time.Hour)),
+				Window7dStart: nil,
+			},
+			want5h: 0,
+			want1d: 10.0,
+			want7d: 50.0,
+		},
+		{
+			name: "zero usage with active windows",
+			key: APIKey{
+				Usage5h:       0,
+				Usage1d:       0,
+				Usage7d:       0,
+				Window5hStart: rateLimitTimePtr(now.Add(-1 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-1 * time.Hour)),
+				Window7dStart: rateLimitTimePtr(now.Add(-1 * time.Hour)),
+			},
+			want5h: 0,
+			want1d: 0,
+			want7d: 0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := tt.key.EffectiveUsage5h(); got != tt.want5h {
+				t.Errorf("EffectiveUsage5h() = %v, want %v", got, tt.want5h)
+			}
+			if got := tt.key.EffectiveUsage1d(); got != tt.want1d {
+				t.Errorf("EffectiveUsage1d() = %v, want %v", got, tt.want1d)
+			}
+			if got := tt.key.EffectiveUsage7d(); got != tt.want7d {
+				t.Errorf("EffectiveUsage7d() = %v, want %v", got, tt.want7d)
+			}
+		})
+	}
+}
+
+func TestAPIKeyRateLimitData_EffectiveUsage(t *testing.T) {
+	now := time.Now()
+
+	tests := []struct {
+		name   string
+		data   APIKeyRateLimitData
+		want5h float64
+		want1d float64
+		want7d float64
+	}{
+		{
+			name: "all windows active",
+			data: APIKeyRateLimitData{
+				Usage5h:       3.0,
+				Usage1d:       8.0,
+				Usage7d:       40.0,
+				Window5hStart: rateLimitTimePtr(now.Add(-2 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-10 * time.Hour)),
+				Window7dStart: rateLimitTimePtr(now.Add(-2 * 24 * time.Hour)),
+			},
+			want5h: 3.0,
+			want1d: 8.0,
+			want7d: 40.0,
+		},
+		{
+			name: "all windows expired",
+			data: APIKeyRateLimitData{
+				Usage5h:       3.0,
+				Usage1d:       8.0,
+				Usage7d:       40.0,
+				Window5hStart: rateLimitTimePtr(now.Add(-10 * time.Hour)),
+				Window1dStart: rateLimitTimePtr(now.Add(-48 * time.Hour)),
+				Window7dStart: rateLimitTimePtr(now.Add(-10 * 24 * time.Hour)),
+			},
+			want5h: 0,
+			want1d: 0,
+			want7d: 0,
+		},
+		{
+			name: "nil window starts return raw usage",
+			data: APIKeyRateLimitData{
+				Usage5h:       3.0,
+				Usage1d:       8.0,
+				Usage7d:       40.0,
+				Window5hStart: nil,
+				Window1dStart: nil,
+				Window7dStart: nil,
+			},
+			want5h: 3.0,
+			want1d: 8.0,
+			want7d: 40.0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := tt.data.EffectiveUsage5h(); got != tt.want5h {
+				t.Errorf("EffectiveUsage5h() = %v, want %v", got, tt.want5h)
+			}
+			if got := tt.data.EffectiveUsage1d(); got != tt.want1d {
+				t.Errorf("EffectiveUsage1d() = %v, want %v", got, tt.want1d)
+			}
+			if got := tt.data.EffectiveUsage7d(); got != tt.want7d {
+				t.Errorf("EffectiveUsage7d() = %v, want %v", got, tt.want7d)
+			}
+		})
+	}
+}
+
+func rateLimitTimePtr(t time.Time) *time.Time {
+	return &t
+}

backend/internal/service/api_key_service.go

@@ -86,6 +86,30 @@ type APIKeyRateLimitData struct {
 	Window7dStart *time.Time
 }
 
+// EffectiveUsage5h returns the 5h window usage, or 0 if the window has expired.
+func (d *APIKeyRateLimitData) EffectiveUsage5h() float64 {
+	if IsWindowExpired(d.Window5hStart, RateLimitWindow5h) {
+		return 0
+	}
+	return d.Usage5h
+}
+
+// EffectiveUsage1d returns the 1d window usage, or 0 if the window has expired.
+func (d *APIKeyRateLimitData) EffectiveUsage1d() float64 {
+	if IsWindowExpired(d.Window1dStart, RateLimitWindow1d) {
+		return 0
+	}
+	return d.Usage1d
+}
+
+// EffectiveUsage7d returns the 7d window usage, or 0 if the window has expired.
+func (d *APIKeyRateLimitData) EffectiveUsage7d() float64 {
+	if IsWindowExpired(d.Window7dStart, RateLimitWindow7d) {
+		return 0
+	}
+	return d.Usage7d
+}
+
 // APIKeyCache defines cache operations for API key service
 type APIKeyCache interface {
 	GetCreateAttemptCount(ctx context.Context, userID int64) (int, error)

backend/internal/service/billing_cache_service.go

@@ -565,15 +565,15 @@ func (s *BillingCacheService) evaluateRateLimits(ctx context.Context, apiKey *AP
 	needsReset := false
 
 	// Reset expired windows in-memory for check purposes
-	if w5h != nil && time.Since(*w5h) >= 5*time.Hour {
+	if IsWindowExpired(w5h, RateLimitWindow5h) {
 		usage5h = 0
 		needsReset = true
 	}
-	if w1d != nil && time.Since(*w1d) >= 24*time.Hour {
+	if IsWindowExpired(w1d, RateLimitWindow1d) {
 		usage1d = 0
 		needsReset = true
 	}
-	if w7d != nil && time.Since(*w7d) >= 7*24*time.Hour {
+	if IsWindowExpired(w7d, RateLimitWindow7d) {
 		usage7d = 0
 		needsReset = true
 	}
@@ -589,12 +589,16 @@ func (s *BillingCacheService) evaluateRateLimits(ctx context.Context, apiKey *AP
 				if loader, ok := s.apiKeyRateLimitLoader.(interface {
 					ResetRateLimitWindows(ctx context.Context, id int64) error
 				}); ok {
-					_ = loader.ResetRateLimitWindows(resetCtx, keyID)
+					if err := loader.ResetRateLimitWindows(resetCtx, keyID); err != nil {
+						logger.LegacyPrintf("service.billing_cache", "Warning: reset rate limit windows failed for api key %d: %v", keyID, err)
+					}
 				}
 			}
 			// Invalidate cache so next request loads fresh data
 			if s.cache != nil {
-				_ = s.cache.InvalidateAPIKeyRateLimit(resetCtx, keyID)
+				if err := s.cache.InvalidateAPIKeyRateLimit(resetCtx, keyID); err != nil {
+					logger.LegacyPrintf("service.billing_cache", "Warning: invalidate rate limit cache failed for api key %d: %v", keyID, err)
+				}
 			}
 		}()
 	}

backend/internal/service/gateway_service.go

@@ -501,33 +501,34 @@ func (s *GatewayService) TempUnscheduleRetryableError(ctx context.Context, accou
 
 // GatewayService handles API gateway operations
 type GatewayService struct {
-	accountRepo          AccountRepository
-	groupRepo            GroupRepository
-	usageLogRepo         UsageLogRepository
-	userRepo             UserRepository
-	userSubRepo          UserSubscriptionRepository
-	userGroupRateRepo    UserGroupRateRepository
-	cache                GatewayCache
-	digestStore          *DigestSessionStore
-	cfg                  *config.Config
-	schedulerSnapshot    *SchedulerSnapshotService
-	billingService       *BillingService
-	rateLimitService     *RateLimitService
-	billingCacheService  *BillingCacheService
-	identityService      *IdentityService
-	httpUpstream         HTTPUpstream
-	deferredService      *DeferredService
-	concurrencyService   *ConcurrencyService
-	claudeTokenProvider  *ClaudeTokenProvider
-	sessionLimitCache    SessionLimitCache // 会话数量限制缓存（仅 Anthropic OAuth/SetupToken）
-	rpmCache             RPMCache          // RPM 计数缓存（仅 Anthropic OAuth/SetupToken）
-	userGroupRateCache   *gocache.Cache
-	userGroupRateSF      singleflight.Group
-	modelsListCache      *gocache.Cache
-	modelsListCacheTTL   time.Duration
-	responseHeaderFilter *responseheaders.CompiledHeaderFilter
-	debugModelRouting    atomic.Bool
-	debugClaudeMimic     atomic.Bool
+	accountRepo           AccountRepository
+	groupRepo             GroupRepository
+	usageLogRepo          UsageLogRepository
+	userRepo              UserRepository
+	userSubRepo           UserSubscriptionRepository
+	userGroupRateRepo     UserGroupRateRepository
+	cache                 GatewayCache
+	digestStore           *DigestSessionStore
+	cfg                   *config.Config
+	schedulerSnapshot     *SchedulerSnapshotService
+	billingService        *BillingService
+	rateLimitService      *RateLimitService
+	billingCacheService   *BillingCacheService
+	identityService       *IdentityService
+	httpUpstream          HTTPUpstream
+	deferredService       *DeferredService
+	concurrencyService    *ConcurrencyService
+	claudeTokenProvider   *ClaudeTokenProvider
+	sessionLimitCache     SessionLimitCache // 会话数量限制缓存（仅 Anthropic OAuth/SetupToken）
+	rpmCache              RPMCache          // RPM 计数缓存（仅 Anthropic OAuth/SetupToken）
+	userGroupRateResolver *userGroupRateResolver
+	userGroupRateCache    *gocache.Cache
+	userGroupRateSF       singleflight.Group
+	modelsListCache       *gocache.Cache
+	modelsListCacheTTL    time.Duration
+	responseHeaderFilter  *responseheaders.CompiledHeaderFilter
+	debugModelRouting     atomic.Bool
+	debugClaudeMimic      atomic.Bool
 }
 
 // NewGatewayService creates a new GatewayService
@@ -582,6 +583,13 @@ func NewGatewayService(
 		modelsListCacheTTL:   modelsListTTL,
 		responseHeaderFilter: compileResponseHeaderFilter(cfg),
 	}
+	svc.userGroupRateResolver = newUserGroupRateResolver(
+		userGroupRateRepo,
+		svc.userGroupRateCache,
+		userGroupRateTTL,
+		&svc.userGroupRateSF,
+		"service.gateway",
+	)
 	svc.debugModelRouting.Store(parseDebugEnvBool(os.Getenv("SUB2API_DEBUG_MODEL_ROUTING")))
 	svc.debugClaudeMimic.Store(parseDebugEnvBool(os.Getenv("SUB2API_DEBUG_CLAUDE_MIMIC")))
 	return svc
@@ -3320,6 +3328,10 @@ func (s *GatewayService) isModelSupportedByAccount(account *Account, requestedMo
 	if account.Platform == PlatformSora {
 		return s.isSoraModelSupportedByAccount(account, requestedModel)
 	}
+	// OpenAI 透传模式：仅替换认证，允许所有模型
+	if account.Platform == PlatformOpenAI && account.IsOpenAIPassthroughEnabled() {
+		return true
+	}
 	// OAuth/SetupToken 账号使用 Anthropic 标准映射（短ID → 长ID）
 	if account.Platform == PlatformAnthropic && account.Type != AccountTypeAPIKey {
 		requestedModel = claude.NormalizeModelID(requestedModel)
@@ -5417,6 +5429,11 @@ func extractUpstreamErrorMessage(body []byte) string {
 		return m
 	}
 
+	// ChatGPT 内部 API 风格：{"detail":"..."}
+	if d := gjson.GetBytes(body, "detail").String(); strings.TrimSpace(d) != "" {
+		return d
+	}
+
 	// 兜底：尝试顶层 message
 	return gjson.GetBytes(body, "message").String()
 }
@@ -6332,63 +6349,20 @@ func (s *GatewayService) replaceModelInResponseBody(body []byte, fromModel, toMo
 }
 
 func (s *GatewayService) getUserGroupRateMultiplier(ctx context.Context, userID, groupID int64, groupDefaultMultiplier float64) float64 {
-	if s == nil || userID <= 0 || groupID <= 0 {
+	if s == nil {
 		return groupDefaultMultiplier
 	}
-
-	key := fmt.Sprintf("%d:%d", userID, groupID)
-	if s.userGroupRateCache != nil {
-		if cached, ok := s.userGroupRateCache.Get(key); ok {
-			if multiplier, castOK := cached.(float64); castOK {
-				userGroupRateCacheHitTotal.Add(1)
-				return multiplier
-			}
-		}
+	resolver := s.userGroupRateResolver
+	if resolver == nil {
+		resolver = newUserGroupRateResolver(
+			s.userGroupRateRepo,
+			s.userGroupRateCache,
+			resolveUserGroupRateCacheTTL(s.cfg),
+			&s.userGroupRateSF,
+			"service.gateway",
+		)
 	}
-	if s.userGroupRateRepo == nil {
-		return groupDefaultMultiplier
-	}
-	userGroupRateCacheMissTotal.Add(1)
-
-	value, err, shared := s.userGroupRateSF.Do(key, func() (any, error) {
-		if s.userGroupRateCache != nil {
-			if cached, ok := s.userGroupRateCache.Get(key); ok {
-				if multiplier, castOK := cached.(float64); castOK {
-					userGroupRateCacheHitTotal.Add(1)
-					return multiplier, nil
-				}
-			}
-		}
-
-		userGroupRateCacheLoadTotal.Add(1)
-		userRate, repoErr := s.userGroupRateRepo.GetByUserAndGroup(ctx, userID, groupID)
-		if repoErr != nil {
-			return nil, repoErr
-		}
-		multiplier := groupDefaultMultiplier
-		if userRate != nil {
-			multiplier = *userRate
-		}
-		if s.userGroupRateCache != nil {
-			s.userGroupRateCache.Set(key, multiplier, resolveUserGroupRateCacheTTL(s.cfg))
-		}
-		return multiplier, nil
-	})
-	if shared {
-		userGroupRateCacheSFSharedTotal.Add(1)
-	}
-	if err != nil {
-		userGroupRateCacheFallbackTotal.Add(1)
-		logger.LegacyPrintf("service.gateway", "get user group rate failed, fallback to group default: user=%d group=%d err=%v", userID, groupID, err)
-		return groupDefaultMultiplier
-	}
-
-	multiplier, ok := value.(float64)
-	if !ok {
-		userGroupRateCacheFallbackTotal.Add(1)
-		return groupDefaultMultiplier
-	}
-	return multiplier
+	return resolver.Resolve(ctx, userID, groupID, groupDefaultMultiplier)
 }
 
 // RecordUsageInput 记录使用量的输入参数

backend/internal/service/group.go

@@ -57,6 +57,10 @@ type Group struct {
 	// 分组排序
 	SortOrder int
 
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch bool
+	DefaultMappedModel    string
+
 	CreatedAt time.Time
 	UpdatedAt time.Time
 

backend/internal/service/openai_codex_transform.go

@@ -1,13 +1,9 @@
 package service
 
 import (
-	_ "embed"
 	"strings"
 )
 
-//go:embed prompts/codex_cli_instructions.md
-var codexCLIInstructions string
-
 var codexModelMap = map[string]string{
 	"gpt-5.4":                    "gpt-5.4",
 	"gpt-5.4-none":               "gpt-5.4",
@@ -77,7 +73,7 @@ type codexTransformResult struct {
 	PromptCacheKey  string
 }
 
-func applyCodexOAuthTransform(reqBody map[string]any, isCodexCLI bool) codexTransformResult {
+func applyCodexOAuthTransform(reqBody map[string]any, isCodexCLI bool, isCompact bool) codexTransformResult {
 	result := codexTransformResult{}
 	// 工具续链需求会影响存储策略与 input 过滤逻辑。
 	needsToolContinuation := NeedsToolContinuation(reqBody)
@@ -95,15 +91,26 @@ func applyCodexOAuthTransform(reqBody map[string]any, isCodexCLI bool) codexTran
 		result.NormalizedModel = normalizedModel
 	}
 
-	// OAuth 走 ChatGPT internal API 时，store 必须为 false；显式 true 也会强制覆盖。
-	// 避免上游返回 "Store must be set to false"。
-	if v, ok := reqBody["store"].(bool); !ok || v {
-		reqBody["store"] = false
-		result.Modified = true
-	}
-	if v, ok := reqBody["stream"].(bool); !ok || !v {
-		reqBody["stream"] = true
-		result.Modified = true
+	if isCompact {
+		if _, ok := reqBody["store"]; ok {
+			delete(reqBody, "store")
+			result.Modified = true
+		}
+		if _, ok := reqBody["stream"]; ok {
+			delete(reqBody, "stream")
+			result.Modified = true
+		}
+	} else {
+		// OAuth 走 ChatGPT internal API 时，store 必须为 false；显式 true 也会强制覆盖。
+		// 避免上游返回 "Store must be set to false"。
+		if v, ok := reqBody["store"].(bool); !ok || v {
+			reqBody["store"] = false
+			result.Modified = true
+		}
+		if v, ok := reqBody["stream"].(bool); !ok || !v {
+			reqBody["stream"] = true
+			result.Modified = true
+		}
 	}
 
 	// Strip parameters unsupported by codex models via the Responses API.
@@ -219,72 +226,13 @@ func getNormalizedCodexModel(modelID string) string {
 	return ""
 }
 
-func getOpenCodeCodexHeader() string {
-	// 兼容保留：历史上这里会从 opencode 仓库拉取 codex_header.txt。
-	// 现在我们与 Codex CLI 一致，直接使用仓库内置的 instructions，避免读写缓存与外网依赖。
-	return getCodexCLIInstructions()
-}
-
-func getCodexCLIInstructions() string {
-	return codexCLIInstructions
-}
-
-func GetOpenCodeInstructions() string {
-	return getOpenCodeCodexHeader()
-}
-
-// GetCodexCLIInstructions 返回内置的 Codex CLI 指令内容。
-func GetCodexCLIInstructions() string {
-	return getCodexCLIInstructions()
-}
-
-// applyInstructions 处理 instructions 字段
-// isCodexCLI=true: 仅补充缺失的 instructions（使用内置 Codex CLI 指令）
-// isCodexCLI=false: 优先使用内置 Codex CLI 指令覆盖
+// applyInstructions 处理 instructions 字段：仅在 instructions 为空时填充默认值。
 func applyInstructions(reqBody map[string]any, isCodexCLI bool) bool {
-	if isCodexCLI {
-		return applyCodexCLIInstructions(reqBody)
-	}
-	return applyOpenCodeInstructions(reqBody)
-}
-
-// applyCodexCLIInstructions 为 Codex CLI 请求补充缺失的 instructions
-// 仅在 instructions 为空时添加内置 Codex CLI 指令（不依赖 opencode 缓存/回源）
-func applyCodexCLIInstructions(reqBody map[string]any) bool {
 	if !isInstructionsEmpty(reqBody) {
-		return false // 已有有效 instructions，不修改
+		return false
 	}
-
-	instructions := strings.TrimSpace(getCodexCLIInstructions())
-	if instructions != "" {
-		reqBody["instructions"] = instructions
-		return true
-	}
-
-	return false
-}
-
-// applyOpenCodeInstructions 为非 Codex CLI 请求应用内置 Codex CLI 指令（兼容历史函数名）
-// 优先使用内置 Codex CLI 指令覆盖
-func applyOpenCodeInstructions(reqBody map[string]any) bool {
-	instructions := strings.TrimSpace(getOpenCodeCodexHeader())
-	existingInstructions, _ := reqBody["instructions"].(string)
-	existingInstructions = strings.TrimSpace(existingInstructions)
-
-	if instructions != "" {
-		if existingInstructions != instructions {
-			reqBody["instructions"] = instructions
-			return true
-		}
-	} else if existingInstructions == "" {
-		codexInstructions := strings.TrimSpace(getCodexCLIInstructions())
-		if codexInstructions != "" {
-			reqBody["instructions"] = codexInstructions
-			return true
-		}
-	}
-
-	return false
+	reqBody["instructions"] = "You are a helpful coding assistant."
+	return true
 }
 
 // isInstructionsEmpty 检查 instructions 字段是否为空

backend/internal/service/openai_codex_transform_test.go

@@ -18,7 +18,7 @@ func TestApplyCodexOAuthTransform_ToolContinuationPreservesInput(t *testing.T) {
 		"tool_choice": "auto",
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	// 未显式设置 store=true，默认为 false。
 	store, ok := reqBody["store"].(bool)
@@ -53,7 +53,7 @@ func TestApplyCodexOAuthTransform_ExplicitStoreFalsePreserved(t *testing.T) {
 		"tool_choice": "auto",
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	store, ok := reqBody["store"].(bool)
 	require.True(t, ok)
@@ -72,13 +72,29 @@ func TestApplyCodexOAuthTransform_ExplicitStoreTrueForcedFalse(t *testing.T) {
 		"tool_choice": "auto",
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	store, ok := reqBody["store"].(bool)
 	require.True(t, ok)
 	require.False(t, store)
 }
 
+func TestApplyCodexOAuthTransform_CompactForcesNonStreaming(t *testing.T) {
+	reqBody := map[string]any{
+		"model":  "gpt-5.1-codex",
+		"store":  true,
+		"stream": true,
+	}
+
+	result := applyCodexOAuthTransform(reqBody, true, true)
+
+	_, hasStore := reqBody["store"]
+	require.False(t, hasStore)
+	_, hasStream := reqBody["stream"]
+	require.False(t, hasStream)
+	require.True(t, result.Modified)
+}
+
 func TestApplyCodexOAuthTransform_NonContinuationDefaultsStoreFalseAndStripsIDs(t *testing.T) {
 	// 非续链场景：未设置 store 时默认 false，并移除 input 中的 id。
 
@@ -89,7 +105,7 @@ func TestApplyCodexOAuthTransform_NonContinuationDefaultsStoreFalseAndStripsIDs(
 		},
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	store, ok := reqBody["store"].(bool)
 	require.True(t, ok)
@@ -138,7 +154,7 @@ func TestApplyCodexOAuthTransform_NormalizeCodexTools_PreservesResponsesFunction
 		},
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	tools, ok := reqBody["tools"].([]any)
 	require.True(t, ok)
@@ -158,7 +174,7 @@ func TestApplyCodexOAuthTransform_EmptyInput(t *testing.T) {
 		"input": []any{},
 	}
 
-	applyCodexOAuthTransform(reqBody, false)
+	applyCodexOAuthTransform(reqBody, false, false)
 
 	input, ok := reqBody["input"].([]any)
 	require.True(t, ok)
@@ -193,7 +209,7 @@ func TestApplyCodexOAuthTransform_CodexCLI_PreservesExistingInstructions(t *test
 		"instructions": "existing instructions",
 	}
 
-	result := applyCodexOAuthTransform(reqBody, true) // isCodexCLI=true
+	result := applyCodexOAuthTransform(reqBody, true, false) // isCodexCLI=true
 
 	instructions, ok := reqBody["instructions"].(string)
 	require.True(t, ok)
@@ -210,7 +226,7 @@ func TestApplyCodexOAuthTransform_CodexCLI_SuppliesDefaultWhenEmpty(t *testing.T
 		// 没有 instructions 字段
 	}
 
-	result := applyCodexOAuthTransform(reqBody, true) // isCodexCLI=true
+	result := applyCodexOAuthTransform(reqBody, true, false) // isCodexCLI=true
 
 	instructions, ok := reqBody["instructions"].(string)
 	require.True(t, ok)
@@ -218,20 +234,19 @@ func TestApplyCodexOAuthTransform_CodexCLI_SuppliesDefaultWhenEmpty(t *testing.T
 	require.True(t, result.Modified)
 }
 
-func TestApplyCodexOAuthTransform_NonCodexCLI_OverridesInstructions(t *testing.T) {
-	// 非 Codex CLI 场景：使用内置 Codex CLI 指令覆盖
+func TestApplyCodexOAuthTransform_NonCodexCLI_PreservesExistingInstructions(t *testing.T) {
+	// 非 Codex CLI 场景：已有 instructions 时保留客户端的值，不再覆盖
 
 	reqBody := map[string]any{
 		"model":        "gpt-5.1",
 		"instructions": "old instructions",
 	}
 
-	result := applyCodexOAuthTransform(reqBody, false) // isCodexCLI=false
+	applyCodexOAuthTransform(reqBody, false, false) // isCodexCLI=false
 
 	instructions, ok := reqBody["instructions"].(string)
 	require.True(t, ok)
-	require.NotEqual(t, "old instructions", instructions)
-	require.True(t, result.Modified)
+	require.Equal(t, "old instructions", instructions)
 }
 
 func TestIsInstructionsEmpty(t *testing.T) {

backend/internal/service/openai_gateway_messages.go

@@ -0,0 +1,416 @@
+package service
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
+	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
+)
+
+// ForwardAsAnthropic accepts an Anthropic Messages request body, converts it
+// to OpenAI Responses API format, forwards to the OpenAI upstream, and converts
+// the response back to Anthropic Messages format. This enables Claude Code
+// clients to access OpenAI models through the standard /v1/messages endpoint.
+func (s *OpenAIGatewayService) ForwardAsAnthropic(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	promptCacheKey string,
+	defaultMappedModel string,
+) (*OpenAIForwardResult, error) {
+	startTime := time.Now()
+
+	// 1. Parse Anthropic request
+	var anthropicReq apicompat.AnthropicRequest
+	if err := json.Unmarshal(body, &anthropicReq); err != nil {
+		return nil, fmt.Errorf("parse anthropic request: %w", err)
+	}
+	originalModel := anthropicReq.Model
+	isStream := anthropicReq.Stream
+
+	// 2. Convert Anthropic → Responses
+	responsesReq, err := apicompat.AnthropicToResponses(&anthropicReq)
+	if err != nil {
+		return nil, fmt.Errorf("convert anthropic to responses: %w", err)
+	}
+
+	// 3. Model mapping
+	mappedModel := account.GetMappedModel(originalModel)
+	// 分组级降级：账号未映射时使用分组默认映射模型
+	if mappedModel == originalModel && defaultMappedModel != "" {
+		mappedModel = defaultMappedModel
+	}
+	responsesReq.Model = mappedModel
+
+	logger.L().Debug("openai messages: model mapping applied",
+		zap.Int64("account_id", account.ID),
+		zap.String("original_model", originalModel),
+		zap.String("mapped_model", mappedModel),
+		zap.Bool("stream", isStream),
+	)
+
+	// 4. Marshal Responses request body, then apply OAuth codex transform
+	responsesBody, err := json.Marshal(responsesReq)
+	if err != nil {
+		return nil, fmt.Errorf("marshal responses request: %w", err)
+	}
+
+	if account.Type == AccountTypeOAuth {
+		var reqBody map[string]any
+		if err := json.Unmarshal(responsesBody, &reqBody); err != nil {
+			return nil, fmt.Errorf("unmarshal for codex transform: %w", err)
+		}
+		applyCodexOAuthTransform(reqBody, false, false)
+		// OAuth codex transform forces stream=true upstream, so always use
+		// the streaming response handler regardless of what the client asked.
+		isStream = true
+		responsesBody, err = json.Marshal(reqBody)
+		if err != nil {
+			return nil, fmt.Errorf("remarshal after codex transform: %w", err)
+		}
+	}
+
+	// 5. Get access token
+	token, _, err := s.GetAccessToken(ctx, account)
+	if err != nil {
+		return nil, fmt.Errorf("get access token: %w", err)
+	}
+
+	// 6. Build upstream request
+	upstreamReq, err := s.buildUpstreamRequest(ctx, c, account, responsesBody, token, isStream, promptCacheKey, false)
+	if err != nil {
+		return nil, fmt.Errorf("build upstream request: %w", err)
+	}
+
+	// 7. Send request
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		safeErr := sanitizeUpstreamErrorMessage(err.Error())
+		setOpsUpstreamError(c, 0, safeErr, "")
+		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+			Platform:           account.Platform,
+			AccountID:          account.ID,
+			AccountName:        account.Name,
+			UpstreamStatusCode: 0,
+			Kind:               "request_error",
+			Message:            safeErr,
+		})
+		writeAnthropicError(c, http.StatusBadGateway, "api_error", "Upstream request failed")
+		return nil, fmt.Errorf("upstream request failed: %s", safeErr)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	// 8. Handle error response with failover
+	if resp.StatusCode >= 400 {
+		if s.shouldFailoverUpstreamError(resp.StatusCode) {
+			respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+			_ = resp.Body.Close()
+
+			upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(respBody))
+			upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
+			upstreamDetail := ""
+			if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+				maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+				if maxBytes <= 0 {
+					maxBytes = 2048
+				}
+				upstreamDetail = truncateString(string(respBody), maxBytes)
+			}
+			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+				Platform:           account.Platform,
+				AccountID:          account.ID,
+				AccountName:        account.Name,
+				UpstreamStatusCode: resp.StatusCode,
+				UpstreamRequestID:  resp.Header.Get("x-request-id"),
+				Kind:               "failover",
+				Message:            upstreamMsg,
+				Detail:             upstreamDetail,
+			})
+			if s.rateLimitService != nil {
+				s.rateLimitService.HandleUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
+			}
+			return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: respBody}
+		}
+		// Non-failover error: return Anthropic-formatted error to client
+		return s.handleAnthropicErrorResponse(resp, c, account)
+	}
+
+	// 9. Handle normal response
+	var result *OpenAIForwardResult
+	var handleErr error
+	if isStream {
+		result, handleErr = s.handleAnthropicStreamingResponse(resp, c, originalModel, mappedModel, startTime)
+	} else {
+		result, handleErr = s.handleAnthropicNonStreamingResponse(resp, c, originalModel, mappedModel, startTime)
+	}
+
+	// Extract and save Codex usage snapshot from response headers (for OAuth accounts)
+	if handleErr == nil && account.Type == AccountTypeOAuth {
+		if snapshot := ParseCodexRateLimitHeaders(resp.Header); snapshot != nil {
+			s.updateCodexUsageSnapshot(ctx, account.ID, snapshot)
+		}
+	}
+
+	return result, handleErr
+}
+
+// handleAnthropicErrorResponse reads an upstream error and returns it in
+// Anthropic error format.
+func (s *OpenAIGatewayService) handleAnthropicErrorResponse(
+	resp *http.Response,
+	c *gin.Context,
+	account *Account,
+) (*OpenAIForwardResult, error) {
+	body, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+
+	upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(body))
+	if upstreamMsg == "" {
+		upstreamMsg = fmt.Sprintf("Upstream error: %d", resp.StatusCode)
+	}
+	upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
+
+	// Record upstream error details for ops logging
+	upstreamDetail := ""
+	if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+		maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+		if maxBytes <= 0 {
+			maxBytes = 2048
+		}
+		upstreamDetail = truncateString(string(body), maxBytes)
+	}
+	setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail)
+
+	// Apply error passthrough rules (matches handleErrorResponse pattern in openai_gateway_service.go)
+	if status, errType, errMsg, matched := applyErrorPassthroughRule(
+		c, account.Platform, resp.StatusCode, body,
+		http.StatusBadGateway, "api_error", "Upstream request failed",
+	); matched {
+		writeAnthropicError(c, status, errType, errMsg)
+		if upstreamMsg == "" {
+			upstreamMsg = errMsg
+		}
+		if upstreamMsg == "" {
+			return nil, fmt.Errorf("upstream error: %d (passthrough rule matched)", resp.StatusCode)
+		}
+		return nil, fmt.Errorf("upstream error: %d (passthrough rule matched) message=%s", resp.StatusCode, upstreamMsg)
+	}
+
+	errType := "api_error"
+	switch {
+	case resp.StatusCode == 400:
+		errType = "invalid_request_error"
+	case resp.StatusCode == 404:
+		errType = "not_found_error"
+	case resp.StatusCode == 429:
+		errType = "rate_limit_error"
+	case resp.StatusCode >= 500:
+		errType = "api_error"
+	}
+
+	writeAnthropicError(c, resp.StatusCode, errType, upstreamMsg)
+	return nil, fmt.Errorf("upstream error: %d %s", resp.StatusCode, upstreamMsg)
+}
+
+// handleAnthropicNonStreamingResponse reads a Responses API JSON response,
+// converts it to Anthropic Messages format, and writes it to the client.
+func (s *OpenAIGatewayService) handleAnthropicNonStreamingResponse(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	mappedModel string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("read upstream response: %w", err)
+	}
+
+	var responsesResp apicompat.ResponsesResponse
+	if err := json.Unmarshal(respBody, &responsesResp); err != nil {
+		return nil, fmt.Errorf("parse responses response: %w", err)
+	}
+
+	anthropicResp := apicompat.ResponsesToAnthropic(&responsesResp, originalModel)
+
+	var usage OpenAIUsage
+	if responsesResp.Usage != nil {
+		usage = OpenAIUsage{
+			InputTokens:  responsesResp.Usage.InputTokens,
+			OutputTokens: responsesResp.Usage.OutputTokens,
+		}
+		if responsesResp.Usage.InputTokensDetails != nil {
+			usage.CacheReadInputTokens = responsesResp.Usage.InputTokensDetails.CachedTokens
+		}
+	}
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	c.JSON(http.StatusOK, anthropicResp)
+
+	return &OpenAIForwardResult{
+		RequestID:    requestID,
+		Usage:        usage,
+		Model:        originalModel,
+		BillingModel: mappedModel,
+		Stream:       false,
+		Duration:     time.Since(startTime),
+	}, nil
+}
+
+// handleAnthropicStreamingResponse reads Responses SSE events from upstream,
+// converts each to Anthropic SSE events, and writes them to the client.
+func (s *OpenAIGatewayService) handleAnthropicStreamingResponse(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	mappedModel string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.WriteHeader(http.StatusOK)
+
+	state := apicompat.NewResponsesEventToAnthropicState()
+	state.Model = originalModel
+	var usage OpenAIUsage
+	var firstTokenMs *int
+	firstChunk := true
+
+	scanner := bufio.NewScanner(resp.Body)
+	scanner.Buffer(make([]byte, 0, 64*1024), 1024*1024)
+
+	for scanner.Scan() {
+		line := scanner.Text()
+
+		if !strings.HasPrefix(line, "data: ") || line == "data: [DONE]" {
+			continue
+		}
+		payload := line[6:]
+
+		if firstChunk {
+			firstChunk = false
+			ms := int(time.Since(startTime).Milliseconds())
+			firstTokenMs = &ms
+		}
+
+		// Parse the Responses SSE event
+		var event apicompat.ResponsesStreamEvent
+		if err := json.Unmarshal([]byte(payload), &event); err != nil {
+			logger.L().Warn("openai messages stream: failed to parse event",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+			continue
+		}
+
+		// Extract usage from completion events
+		if (event.Type == "response.completed" || event.Type == "response.incomplete" || event.Type == "response.failed") &&
+			event.Response != nil && event.Response.Usage != nil {
+			usage = OpenAIUsage{
+				InputTokens:  event.Response.Usage.InputTokens,
+				OutputTokens: event.Response.Usage.OutputTokens,
+			}
+			if event.Response.Usage.InputTokensDetails != nil {
+				usage.CacheReadInputTokens = event.Response.Usage.InputTokensDetails.CachedTokens
+			}
+		}
+
+		// Convert to Anthropic events
+		events := apicompat.ResponsesEventToAnthropicEvents(&event, state)
+		for _, evt := range events {
+			sse, err := apicompat.ResponsesAnthropicEventToSSE(evt)
+			if err != nil {
+				logger.L().Warn("openai messages stream: failed to marshal event",
+					zap.Error(err),
+					zap.String("request_id", requestID),
+				)
+				continue
+			}
+			if _, err := fmt.Fprint(c.Writer, sse); err != nil {
+				// Client disconnected — return collected usage
+				logger.L().Info("openai messages stream: client disconnected",
+					zap.String("request_id", requestID),
+				)
+				return &OpenAIForwardResult{
+					RequestID:    requestID,
+					Usage:        usage,
+					Model:        originalModel,
+					BillingModel: mappedModel,
+					Stream:       true,
+					Duration:     time.Since(startTime),
+					FirstTokenMs: firstTokenMs,
+				}, nil
+			}
+		}
+		if len(events) > 0 {
+			c.Writer.Flush()
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai messages stream: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	// Ensure the Anthropic stream is properly terminated
+	if finalEvents := apicompat.FinalizeResponsesAnthropicStream(state); len(finalEvents) > 0 {
+		for _, evt := range finalEvents {
+			sse, err := apicompat.ResponsesAnthropicEventToSSE(evt)
+			if err != nil {
+				continue
+			}
+			fmt.Fprint(c.Writer, sse) //nolint:errcheck
+		}
+		c.Writer.Flush()
+	}
+
+	return &OpenAIForwardResult{
+		RequestID:    requestID,
+		Usage:        usage,
+		Model:        originalModel,
+		BillingModel: mappedModel,
+		Stream:       true,
+		Duration:     time.Since(startTime),
+		FirstTokenMs: firstTokenMs,
+	}, nil
+}
+
+// writeAnthropicError writes an error response in Anthropic Messages API format.
+func writeAnthropicError(c *gin.Context, statusCode int, errType, message string) {
+	c.JSON(statusCode, gin.H{
+		"type": "error",
+		"error": gin.H{
+			"type":    errType,
+			"message": message,
+		},
+	})
+}

backend/internal/service/openai_gateway_record_usage_test.go

@@ -0,0 +1,336 @@
+package service
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+type openAIRecordUsageLogRepoStub struct {
+	UsageLogRepository
+
+	inserted bool
+	err      error
+	calls    int
+	lastLog  *UsageLog
+}
+
+func (s *openAIRecordUsageLogRepoStub) Create(ctx context.Context, log *UsageLog) (bool, error) {
+	s.calls++
+	s.lastLog = log
+	return s.inserted, s.err
+}
+
+type openAIRecordUsageUserRepoStub struct {
+	UserRepository
+
+	deductCalls int
+	deductErr   error
+	lastAmount  float64
+}
+
+func (s *openAIRecordUsageUserRepoStub) DeductBalance(ctx context.Context, id int64, amount float64) error {
+	s.deductCalls++
+	s.lastAmount = amount
+	return s.deductErr
+}
+
+type openAIRecordUsageSubRepoStub struct {
+	UserSubscriptionRepository
+
+	incrementCalls int
+	incrementErr   error
+}
+
+func (s *openAIRecordUsageSubRepoStub) IncrementUsage(ctx context.Context, id int64, costUSD float64) error {
+	s.incrementCalls++
+	return s.incrementErr
+}
+
+type openAIRecordUsageAPIKeyQuotaStub struct {
+	quotaCalls     int
+	rateLimitCalls int
+	err            error
+	lastAmount     float64
+}
+
+func (s *openAIRecordUsageAPIKeyQuotaStub) UpdateQuotaUsed(ctx context.Context, apiKeyID int64, cost float64) error {
+	s.quotaCalls++
+	s.lastAmount = cost
+	return s.err
+}
+
+func (s *openAIRecordUsageAPIKeyQuotaStub) UpdateRateLimitUsage(ctx context.Context, apiKeyID int64, cost float64) error {
+	s.rateLimitCalls++
+	s.lastAmount = cost
+	return s.err
+}
+
+type openAIUserGroupRateRepoStub struct {
+	UserGroupRateRepository
+
+	rate  *float64
+	err   error
+	calls int
+}
+
+func (s *openAIUserGroupRateRepoStub) GetByUserAndGroup(ctx context.Context, userID, groupID int64) (*float64, error) {
+	s.calls++
+	if s.err != nil {
+		return nil, s.err
+	}
+	return s.rate, nil
+}
+
+func i64p(v int64) *int64 {
+	return &v
+}
+
+func newOpenAIRecordUsageServiceForTest(usageRepo UsageLogRepository, userRepo UserRepository, subRepo UserSubscriptionRepository, rateRepo UserGroupRateRepository) *OpenAIGatewayService {
+	cfg := &config.Config{}
+	cfg.Default.RateMultiplier = 1.1
+
+	return &OpenAIGatewayService{
+		usageLogRepo:        usageRepo,
+		userRepo:            userRepo,
+		userSubRepo:         subRepo,
+		cfg:                 cfg,
+		billingService:      NewBillingService(cfg, nil),
+		billingCacheService: &BillingCacheService{},
+		deferredService:     &DeferredService{},
+		userGroupRateResolver: newUserGroupRateResolver(
+			rateRepo,
+			nil,
+			resolveUserGroupRateCacheTTL(cfg),
+			nil,
+			"service.openai_gateway.test",
+		),
+	}
+}
+
+func expectedOpenAICost(t *testing.T, svc *OpenAIGatewayService, model string, usage OpenAIUsage, multiplier float64) *CostBreakdown {
+	t.Helper()
+
+	cost, err := svc.billingService.CalculateCost(model, UsageTokens{
+		InputTokens:         max(usage.InputTokens-usage.CacheReadInputTokens, 0),
+		OutputTokens:        usage.OutputTokens,
+		CacheCreationTokens: usage.CacheCreationInputTokens,
+		CacheReadTokens:     usage.CacheReadInputTokens,
+	}, multiplier)
+	require.NoError(t, err)
+	return cost
+}
+
+func max(a, b int) int {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func TestOpenAIGatewayServiceRecordUsage_UsesUserSpecificGroupRate(t *testing.T) {
+	groupID := int64(11)
+	groupRate := 1.4
+	userRate := 1.8
+	usage := OpenAIUsage{InputTokens: 15, OutputTokens: 4, CacheReadInputTokens: 3}
+
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	rateRepo := &openAIUserGroupRateRepoStub{rate: &userRate}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, rateRepo)
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_user_group_rate",
+			Usage:     usage,
+			Model:     "gpt-5.1",
+			Duration:  time.Second,
+		},
+		APIKey: &APIKey{
+			ID:      1001,
+			GroupID: i64p(groupID),
+			Group: &Group{
+				ID:             groupID,
+				RateMultiplier: groupRate,
+			},
+		},
+		User:    &User{ID: 2001},
+		Account: &Account{ID: 3001},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, rateRepo.calls)
+	require.NotNil(t, usageRepo.lastLog)
+	require.Equal(t, userRate, usageRepo.lastLog.RateMultiplier)
+	require.Equal(t, 12, usageRepo.lastLog.InputTokens)
+	require.Equal(t, 3, usageRepo.lastLog.CacheReadTokens)
+
+	expected := expectedOpenAICost(t, svc, "gpt-5.1", usage, userRate)
+	require.InDelta(t, expected.ActualCost, usageRepo.lastLog.ActualCost, 1e-12)
+	require.InDelta(t, expected.ActualCost, userRepo.lastAmount, 1e-12)
+	require.Equal(t, 1, userRepo.deductCalls)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_FallsBackToGroupDefaultRateOnResolverError(t *testing.T) {
+	groupID := int64(12)
+	groupRate := 1.6
+	usage := OpenAIUsage{InputTokens: 10, OutputTokens: 5, CacheReadInputTokens: 2}
+
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	rateRepo := &openAIUserGroupRateRepoStub{err: errors.New("db unavailable")}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, rateRepo)
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_group_default_on_error",
+			Usage:     usage,
+			Model:     "gpt-5.1",
+			Duration:  time.Second,
+		},
+		APIKey: &APIKey{
+			ID:      1002,
+			GroupID: i64p(groupID),
+			Group: &Group{
+				ID:             groupID,
+				RateMultiplier: groupRate,
+			},
+		},
+		User:    &User{ID: 2002},
+		Account: &Account{ID: 3002},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, rateRepo.calls)
+	require.NotNil(t, usageRepo.lastLog)
+	require.Equal(t, groupRate, usageRepo.lastLog.RateMultiplier)
+
+	expected := expectedOpenAICost(t, svc, "gpt-5.1", usage, groupRate)
+	require.InDelta(t, expected.ActualCost, userRepo.lastAmount, 1e-12)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_FallsBackToGroupDefaultRateWhenResolverMissing(t *testing.T) {
+	groupID := int64(13)
+	groupRate := 1.25
+	usage := OpenAIUsage{InputTokens: 9, OutputTokens: 4, CacheReadInputTokens: 1}
+
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, nil)
+	svc.userGroupRateResolver = nil
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_group_default_nil_resolver",
+			Usage:     usage,
+			Model:     "gpt-5.1",
+			Duration:  time.Second,
+		},
+		APIKey: &APIKey{
+			ID:      1003,
+			GroupID: i64p(groupID),
+			Group: &Group{
+				ID:             groupID,
+				RateMultiplier: groupRate,
+			},
+		},
+		User:    &User{ID: 2003},
+		Account: &Account{ID: 3003},
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, usageRepo.lastLog)
+	require.Equal(t, groupRate, usageRepo.lastLog.RateMultiplier)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_DuplicateUsageLogSkipsBilling(t *testing.T) {
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: false}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, nil)
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_duplicate",
+			Usage: OpenAIUsage{
+				InputTokens:  8,
+				OutputTokens: 4,
+			},
+			Model:    "gpt-5.1",
+			Duration: time.Second,
+		},
+		APIKey:  &APIKey{ID: 1004},
+		User:    &User{ID: 2004},
+		Account: &Account{ID: 3004},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, usageRepo.calls)
+	require.Equal(t, 0, userRepo.deductCalls)
+	require.Equal(t, 0, subRepo.incrementCalls)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_UpdatesAPIKeyQuotaWhenConfigured(t *testing.T) {
+	usage := OpenAIUsage{InputTokens: 10, OutputTokens: 6, CacheReadInputTokens: 2}
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	quotaSvc := &openAIRecordUsageAPIKeyQuotaStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, nil)
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_quota_update",
+			Usage:     usage,
+			Model:     "gpt-5.1",
+			Duration:  time.Second,
+		},
+		APIKey: &APIKey{
+			ID:    1005,
+			Quota: 100,
+		},
+		User:          &User{ID: 2005},
+		Account:       &Account{ID: 3005},
+		APIKeyService: quotaSvc,
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, quotaSvc.quotaCalls)
+	require.Equal(t, 0, quotaSvc.rateLimitCalls)
+	expected := expectedOpenAICost(t, svc, "gpt-5.1", usage, 1.1)
+	require.InDelta(t, expected.ActualCost, quotaSvc.lastAmount, 1e-12)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_ClampsActualInputTokensToZero(t *testing.T) {
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	subRepo := &openAIRecordUsageSubRepoStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, subRepo, nil)
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "resp_clamp_actual_input",
+			Usage: OpenAIUsage{
+				InputTokens:          2,
+				OutputTokens:         1,
+				CacheReadInputTokens: 5,
+			},
+			Model:    "gpt-5.1",
+			Duration: time.Second,
+		},
+		APIKey:  &APIKey{ID: 1006},
+		User:    &User{ID: 2006},
+		Account: &Account{ID: 3006},
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, usageRepo.lastLog)
+	require.Equal(t, 0, usageRepo.lastLog.InputTokens)
+}

backend/internal/service/openai_gateway_service.go

@@ -25,6 +25,7 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/Wei-Shaw/sub2api/internal/util/urlvalidator"
 	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 	"go.uber.org/zap"
@@ -49,6 +50,8 @@ const (
 	openAIWSRetryBackoffInitialDefault = 120 * time.Millisecond
 	openAIWSRetryBackoffMaxDefault     = 2 * time.Second
 	openAIWSRetryJitterRatioDefault    = 0.2
+	openAICompactSessionSeedKey        = "openai_compact_session_seed"
+	codexCLIVersion                    = "0.104.0"
 )
 
 // OpenAI allowed headers whitelist (for non-passthrough).
@@ -204,12 +207,18 @@ type OpenAIUsage struct {
 type OpenAIForwardResult struct {
 	RequestID string
 	Usage     OpenAIUsage
-	Model     string
+	Model     string // 原始模型（用于响应和日志显示）
+	// BillingModel is the model used for cost calculation.
+	// When non-empty, CalculateCost uses this instead of Model.
+	// This is set by the Anthropic Messages conversion path where
+	// the mapped upstream model differs from the client-facing model.
+	BillingModel string
 	// ReasoningEffort is extracted from request body (reasoning.effort) or derived from model suffix.
 	// Stored for usage records display; nil means not provided / not applicable.
 	ReasoningEffort *string
 	Stream          bool
 	OpenAIWSMode    bool
+	ResponseHeaders http.Header
 	Duration        time.Duration
 	FirstTokenMs    *int
 }
@@ -245,23 +254,24 @@ type openAIWSRetryMetrics struct {
 
 // OpenAIGatewayService handles OpenAI API gateway operations
 type OpenAIGatewayService struct {
-	accountRepo         AccountRepository
-	usageLogRepo        UsageLogRepository
-	userRepo            UserRepository
-	userSubRepo         UserSubscriptionRepository
-	cache               GatewayCache
-	cfg                 *config.Config
-	codexDetector       CodexClientRestrictionDetector
-	schedulerSnapshot   *SchedulerSnapshotService
-	concurrencyService  *ConcurrencyService
-	billingService      *BillingService
-	rateLimitService    *RateLimitService
-	billingCacheService *BillingCacheService
-	httpUpstream        HTTPUpstream
-	deferredService     *DeferredService
-	openAITokenProvider *OpenAITokenProvider
-	toolCorrector       *CodexToolCorrector
-	openaiWSResolver    OpenAIWSProtocolResolver
+	accountRepo           AccountRepository
+	usageLogRepo          UsageLogRepository
+	userRepo              UserRepository
+	userSubRepo           UserSubscriptionRepository
+	cache                 GatewayCache
+	cfg                   *config.Config
+	codexDetector         CodexClientRestrictionDetector
+	schedulerSnapshot     *SchedulerSnapshotService
+	concurrencyService    *ConcurrencyService
+	billingService        *BillingService
+	rateLimitService      *RateLimitService
+	billingCacheService   *BillingCacheService
+	userGroupRateResolver *userGroupRateResolver
+	httpUpstream          HTTPUpstream
+	deferredService       *DeferredService
+	openAITokenProvider   *OpenAITokenProvider
+	toolCorrector         *CodexToolCorrector
+	openaiWSResolver      OpenAIWSProtocolResolver
 
 	openaiWSPoolOnce              sync.Once
 	openaiWSStateStoreOnce        sync.Once
@@ -284,6 +294,7 @@ func NewOpenAIGatewayService(
 	usageLogRepo UsageLogRepository,
 	userRepo UserRepository,
 	userSubRepo UserSubscriptionRepository,
+	userGroupRateRepo UserGroupRateRepository,
 	cache GatewayCache,
 	cfg *config.Config,
 	schedulerSnapshot *SchedulerSnapshotService,
@@ -296,18 +307,25 @@ func NewOpenAIGatewayService(
 	openAITokenProvider *OpenAITokenProvider,
 ) *OpenAIGatewayService {
 	svc := &OpenAIGatewayService{
-		accountRepo:          accountRepo,
-		usageLogRepo:         usageLogRepo,
-		userRepo:             userRepo,
-		userSubRepo:          userSubRepo,
-		cache:                cache,
-		cfg:                  cfg,
-		codexDetector:        NewOpenAICodexClientRestrictionDetector(cfg),
-		schedulerSnapshot:    schedulerSnapshot,
-		concurrencyService:   concurrencyService,
-		billingService:       billingService,
-		rateLimitService:     rateLimitService,
-		billingCacheService:  billingCacheService,
+		accountRepo:         accountRepo,
+		usageLogRepo:        usageLogRepo,
+		userRepo:            userRepo,
+		userSubRepo:         userSubRepo,
+		cache:               cache,
+		cfg:                 cfg,
+		codexDetector:       NewOpenAICodexClientRestrictionDetector(cfg),
+		schedulerSnapshot:   schedulerSnapshot,
+		concurrencyService:  concurrencyService,
+		billingService:      billingService,
+		rateLimitService:    rateLimitService,
+		billingCacheService: billingCacheService,
+		userGroupRateResolver: newUserGroupRateResolver(
+			userGroupRateRepo,
+			nil,
+			resolveUserGroupRateCacheTTL(cfg),
+			nil,
+			"service.openai_gateway",
+		),
 		httpUpstream:         httpUpstream,
 		deferredService:      deferredService,
 		openAITokenProvider:  openAITokenProvider,
@@ -816,8 +834,10 @@ func logOpenAIInstructionsRequiredDebug(
 	}
 
 	userAgent := ""
+	originator := ""
 	if c != nil {
 		userAgent = strings.TrimSpace(c.GetHeader("User-Agent"))
+		originator = strings.TrimSpace(c.GetHeader("originator"))
 	}
 
 	fields := []zap.Field{
@@ -827,7 +847,7 @@ func logOpenAIInstructionsRequiredDebug(
 		zap.Int("upstream_status_code", upstreamStatusCode),
 		zap.String("upstream_error_message", msg),
 		zap.String("request_user_agent", userAgent),
-		zap.Bool("codex_official_client_match", openai.IsCodexCLIRequest(userAgent)),
+		zap.Bool("codex_official_client_match", openai.IsCodexOfficialClientByHeaders(userAgent, originator)),
 	}
 	fields = appendCodexCLIOnlyRejectedRequestFields(fields, c, requestBody)
 
@@ -888,6 +908,22 @@ func isOpenAIInstructionsRequiredError(upstreamStatusCode int, upstreamMsg strin
 	return false
 }
 
+// ExtractSessionID extracts the raw session ID from headers or body without hashing.
+// Used by ForwardAsAnthropic to pass as prompt_cache_key for upstream cache.
+func (s *OpenAIGatewayService) ExtractSessionID(c *gin.Context, body []byte) string {
+	if c == nil {
+		return ""
+	}
+	sessionID := strings.TrimSpace(c.GetHeader("session_id"))
+	if sessionID == "" {
+		sessionID = strings.TrimSpace(c.GetHeader("conversation_id"))
+	}
+	if sessionID == "" && len(body) > 0 {
+		sessionID = strings.TrimSpace(gjson.GetBytes(body, "prompt_cache_key").String())
+	}
+	return sessionID
+}
+
 // GenerateSessionHash generates a sticky-session hash for OpenAI requests.
 //
 // Priority:
@@ -934,6 +970,18 @@ func (s *OpenAIGatewayService) GenerateSessionHashWithFallback(c *gin.Context, b
 	return currentHash
 }
 
+func resolveOpenAIUpstreamOriginator(c *gin.Context, isOfficialClient bool) string {
+	if c != nil {
+		if originator := strings.TrimSpace(c.GetHeader("originator")); originator != "" {
+			return originator
+		}
+	}
+	if isOfficialClient {
+		return "codex_cli_rs"
+	}
+	return "opencode"
+}
+
 // BindStickySession sets session -> account binding with standard TTL.
 func (s *OpenAIGatewayService) BindStickySession(ctx context.Context, groupID *int64, sessionHash string, accountID int64) error {
 	if sessionHash == "" || accountID <= 0 {
@@ -1455,7 +1503,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 	reqModel, reqStream, promptCacheKey := extractOpenAIRequestMetaFromBody(body)
 	originalModel := reqModel
 
-	isCodexCLI := openai.IsCodexCLIRequest(c.GetHeader("User-Agent")) || (s.cfg != nil && s.cfg.Gateway.ForceCodexCLI)
+	isCodexCLI := openai.IsCodexOfficialClientByHeaders(c.GetHeader("User-Agent"), c.GetHeader("originator")) || (s.cfg != nil && s.cfg.Gateway.ForceCodexCLI)
 	wsDecision := s.getOpenAIWSProtocolResolver().Resolve(account)
 	clientTransport := GetOpenAIClientTransport(c)
 	// 仅允许 WS 入站请求走 WS 上游，避免出现 HTTP -> WS 协议混用。
@@ -1563,13 +1611,11 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		patchDisabled = true
 	}
 
-	// 非透传模式下，保持历史行为：非 Codex CLI 请求在 instructions 为空时注入默认指令。
-	if !isCodexCLI && isInstructionsEmpty(reqBody) {
-		if instructions := strings.TrimSpace(GetOpenCodeInstructions()); instructions != "" {
-			reqBody["instructions"] = instructions
-			bodyModified = true
-			markPatchSet("instructions", instructions)
-		}
+	// 非透传模式下，instructions 为空时注入默认指令。
+	if isInstructionsEmpty(reqBody) {
+		reqBody["instructions"] = "You are a helpful coding assistant."
+		bodyModified = true
+		markPatchSet("instructions", "You are a helpful coding assistant.")
 	}
 
 	// 对所有请求执行模型映射（包含 Codex CLI）。
@@ -1605,7 +1651,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 	}
 
 	if account.Type == AccountTypeOAuth {
-		codexResult := applyCodexOAuthTransform(reqBody, isCodexCLI)
+		codexResult := applyCodexOAuthTransform(reqBody, isCodexCLI, isOpenAIResponsesCompactPath(c))
 		if codexResult.Modified {
 			bodyModified = true
 			disablePatch()
@@ -2037,14 +2083,14 @@ func (s *OpenAIGatewayService) forwardOpenAIPassthrough(
 			return nil, fmt.Errorf("openai passthrough rejected before upstream: %s", rejectReason)
 		}
 
-		normalizedBody, normalized, err := normalizeOpenAIPassthroughOAuthBody(body)
+		normalizedBody, normalized, err := normalizeOpenAIPassthroughOAuthBody(body, isOpenAIResponsesCompactPath(c))
 		if err != nil {
 			return nil, err
 		}
 		if normalized {
 			body = normalizedBody
-			reqStream = true
 		}
+		reqStream = gjson.GetBytes(body, "stream").Bool()
 	}
 
 	logger.LegacyPrintf("service.openai_gateway",
@@ -2209,6 +2255,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 			targetURL = buildOpenAIResponsesURL(validatedURL)
 		}
 	}
+	targetURL = appendOpenAIResponsesRequestPathSuffix(targetURL, openAIResponsesRequestPathSuffix(c))
 
 	req, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(body))
 	if err != nil {
@@ -2242,7 +2289,15 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 		if chatgptAccountID := account.GetChatGPTAccountID(); chatgptAccountID != "" {
 			req.Header.Set("chatgpt-account-id", chatgptAccountID)
 		}
-		if req.Header.Get("accept") == "" {
+		if isOpenAIResponsesCompactPath(c) {
+			req.Header.Set("accept", "application/json")
+			if req.Header.Get("version") == "" {
+				req.Header.Set("version", codexCLIVersion)
+			}
+			if req.Header.Get("session_id") == "" {
+				req.Header.Set("session_id", resolveOpenAICompactSessionID(c))
+			}
+		} else if req.Header.Get("accept") == "" {
 			req.Header.Set("accept", "text/event-stream")
 		}
 		if req.Header.Get("OpenAI-Beta") == "" {
@@ -2589,6 +2644,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequest(ctx context.Context, c *gin.
 	default:
 		targetURL = openaiPlatformAPIURL
 	}
+	targetURL = appendOpenAIResponsesRequestPathSuffix(targetURL, openAIResponsesRequestPathSuffix(c))
 
 	req, err := http.NewRequestWithContext(ctx, "POST", targetURL, bytes.NewReader(body))
 	if err != nil {
@@ -2620,12 +2676,18 @@ func (s *OpenAIGatewayService) buildUpstreamRequest(ctx context.Context, c *gin.
 	}
 	if account.Type == AccountTypeOAuth {
 		req.Header.Set("OpenAI-Beta", "responses=experimental")
-		if isCodexCLI {
-			req.Header.Set("originator", "codex_cli_rs")
+		req.Header.Set("originator", resolveOpenAIUpstreamOriginator(c, isCodexCLI))
+		if isOpenAIResponsesCompactPath(c) {
+			req.Header.Set("accept", "application/json")
+			if req.Header.Get("version") == "" {
+				req.Header.Set("version", codexCLIVersion)
+			}
+			if req.Header.Get("session_id") == "" {
+				req.Header.Set("session_id", resolveOpenAICompactSessionID(c))
+			}
 		} else {
-			req.Header.Set("originator", "opencode")
+			req.Header.Set("accept", "text/event-stream")
 		}
-		req.Header.Set("accept", "text/event-stream")
 		if promptCacheKey != "" {
 			req.Header.Set("conversation_id", promptCacheKey)
 			req.Header.Set("session_id", promptCacheKey)
@@ -3261,6 +3323,14 @@ func (s *OpenAIGatewayService) handleOAuthSSEToJSON(resp *http.Response, c *gin.
 		// Correct tool calls in final response
 		body = s.correctToolCallsInResponseBody(body)
 	} else {
+		terminalType, terminalPayload, terminalOK := extractOpenAISSETerminalEvent(bodyText)
+		if terminalOK && terminalType == "response.failed" {
+			msg := extractOpenAISSEErrorMessage(terminalPayload)
+			if msg == "" {
+				msg = "Upstream compact response failed"
+			}
+			return nil, s.writeOpenAINonStreamingProtocolError(resp, c, msg)
+		}
 		usage = s.parseSSEUsageFromBody(bodyText)
 		if originalModel != mappedModel {
 			bodyText = s.replaceModelInSSEBody(bodyText, mappedModel, originalModel)
@@ -3282,6 +3352,51 @@ func (s *OpenAIGatewayService) handleOAuthSSEToJSON(resp *http.Response, c *gin.
 	return usage, nil
 }
 
+func extractOpenAISSETerminalEvent(body string) (string, []byte, bool) {
+	lines := strings.Split(body, "\n")
+	for _, line := range lines {
+		data, ok := extractOpenAISSEDataLine(line)
+		if !ok || data == "" || data == "[DONE]" {
+			continue
+		}
+		eventType := strings.TrimSpace(gjson.Get(data, "type").String())
+		switch eventType {
+		case "response.completed", "response.done", "response.failed":
+			return eventType, []byte(data), true
+		}
+	}
+	return "", nil, false
+}
+
+func extractOpenAISSEErrorMessage(payload []byte) string {
+	if len(payload) == 0 {
+		return ""
+	}
+	for _, path := range []string{"response.error.message", "error.message", "message"} {
+		if msg := strings.TrimSpace(gjson.GetBytes(payload, path).String()); msg != "" {
+			return sanitizeUpstreamErrorMessage(msg)
+		}
+	}
+	return sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(payload)))
+}
+
+func (s *OpenAIGatewayService) writeOpenAINonStreamingProtocolError(resp *http.Response, c *gin.Context, message string) error {
+	message = sanitizeUpstreamErrorMessage(strings.TrimSpace(message))
+	if message == "" {
+		message = "Upstream returned an invalid non-streaming response"
+	}
+	setOpsUpstreamError(c, http.StatusBadGateway, message, "")
+	responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
+	c.JSON(http.StatusBadGateway, gin.H{
+		"error": gin.H{
+			"type":    "upstream_error",
+			"message": message,
+		},
+	})
+	return fmt.Errorf("non-streaming openai protocol error: %s", message)
+}
+
 func extractCodexFinalResponse(body string) ([]byte, bool) {
 	lines := strings.Split(body, "\n")
 	for _, line := range lines {
@@ -3363,6 +3478,95 @@ func buildOpenAIResponsesURL(base string) string {
 	return normalized + "/v1/responses"
 }
 
+func IsOpenAIResponsesCompactPathForTest(c *gin.Context) bool {
+	return isOpenAIResponsesCompactPath(c)
+}
+
+func OpenAICompactSessionSeedKeyForTest() string {
+	return openAICompactSessionSeedKey
+}
+
+func NormalizeOpenAICompactRequestBodyForTest(body []byte) ([]byte, bool, error) {
+	return normalizeOpenAICompactRequestBody(body)
+}
+
+func isOpenAIResponsesCompactPath(c *gin.Context) bool {
+	suffix := strings.TrimSpace(openAIResponsesRequestPathSuffix(c))
+	return suffix == "/compact" || strings.HasPrefix(suffix, "/compact/")
+}
+
+func normalizeOpenAICompactRequestBody(body []byte) ([]byte, bool, error) {
+	if len(body) == 0 {
+		return body, false, nil
+	}
+
+	normalized := []byte(`{}`)
+	for _, field := range []string{"model", "input", "instructions", "previous_response_id"} {
+		value := gjson.GetBytes(body, field)
+		if !value.Exists() {
+			continue
+		}
+		next, err := sjson.SetRawBytes(normalized, field, []byte(value.Raw))
+		if err != nil {
+			return body, false, fmt.Errorf("normalize compact body %s: %w", field, err)
+		}
+		normalized = next
+	}
+
+	if bytes.Equal(bytes.TrimSpace(body), bytes.TrimSpace(normalized)) {
+		return body, false, nil
+	}
+	return normalized, true, nil
+}
+
+func resolveOpenAICompactSessionID(c *gin.Context) string {
+	if c != nil {
+		if sessionID := strings.TrimSpace(c.GetHeader("session_id")); sessionID != "" {
+			return sessionID
+		}
+		if conversationID := strings.TrimSpace(c.GetHeader("conversation_id")); conversationID != "" {
+			return conversationID
+		}
+		if seed, ok := c.Get(openAICompactSessionSeedKey); ok {
+			if seedStr, ok := seed.(string); ok && strings.TrimSpace(seedStr) != "" {
+				return strings.TrimSpace(seedStr)
+			}
+		}
+	}
+	return uuid.NewString()
+}
+
+func openAIResponsesRequestPathSuffix(c *gin.Context) string {
+	if c == nil || c.Request == nil || c.Request.URL == nil {
+		return ""
+	}
+	normalizedPath := strings.TrimRight(strings.TrimSpace(c.Request.URL.Path), "/")
+	if normalizedPath == "" {
+		return ""
+	}
+	idx := strings.LastIndex(normalizedPath, "/responses")
+	if idx < 0 {
+		return ""
+	}
+	suffix := normalizedPath[idx+len("/responses"):]
+	if suffix == "" || suffix == "/" {
+		return ""
+	}
+	if !strings.HasPrefix(suffix, "/") {
+		return ""
+	}
+	return suffix
+}
+
+func appendOpenAIResponsesRequestPathSuffix(baseURL, suffix string) string {
+	trimmedBase := strings.TrimRight(strings.TrimSpace(baseURL), "/")
+	trimmedSuffix := strings.TrimSpace(suffix)
+	if trimmedBase == "" || trimmedSuffix == "" {
+		return trimmedBase
+	}
+	return trimmedBase + trimmedSuffix
+}
+
 func (s *OpenAIGatewayService) replaceModelInResponseBody(body []byte, fromModel, toModel string) []byte {
 	// 使用 gjson/sjson 精确替换 model 字段，避免全量 JSON 反序列化
 	if m := gjson.GetBytes(body, "model"); m.Exists() && m.Str == fromModel {
@@ -3413,10 +3617,18 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 	// Get rate multiplier
 	multiplier := s.cfg.Default.RateMultiplier
 	if apiKey.GroupID != nil && apiKey.Group != nil {
-		multiplier = apiKey.Group.RateMultiplier
+		resolver := s.userGroupRateResolver
+		if resolver == nil {
+			resolver = newUserGroupRateResolver(nil, nil, resolveUserGroupRateCacheTTL(s.cfg), nil, "service.openai_gateway")
+		}
+		multiplier = resolver.Resolve(ctx, user.ID, *apiKey.GroupID, apiKey.Group.RateMultiplier)
 	}
 
-	cost, err := s.billingService.CalculateCost(result.Model, tokens, multiplier)
+	billingModel := result.Model
+	if result.BillingModel != "" {
+		billingModel = result.BillingModel
+	}
+	cost, err := s.billingService.CalculateCost(billingModel, tokens, multiplier)
 	if err != nil {
 		cost = &CostBreakdown{ActualCost: 0}
 	}
@@ -3436,7 +3648,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 		APIKeyID:              apiKey.ID,
 		AccountID:             account.ID,
 		RequestID:             result.RequestID,
-		Model:                 result.Model,
+		Model:                 billingModel,
 		ReasoningEffort:       result.ReasoningEffort,
 		InputTokens:           actualInputTokens,
 		OutputTokens:          result.Usage.OutputTokens,
@@ -3681,6 +3893,15 @@ func (s *OpenAIGatewayService) updateCodexUsageSnapshot(ctx context.Context, acc
 	}()
 }
 
+func (s *OpenAIGatewayService) UpdateCodexUsageSnapshotFromHeaders(ctx context.Context, accountID int64, headers http.Header) {
+	if accountID <= 0 || headers == nil {
+		return
+	}
+	if snapshot := ParseCodexRateLimitHeaders(headers); snapshot != nil {
+		s.updateCodexUsageSnapshot(ctx, accountID, snapshot)
+	}
+}
+
 func getOpenAIReasoningEffortFromReqBody(reqBody map[string]any) (value string, present bool) {
 	if reqBody == nil {
 		return "", false
@@ -3739,8 +3960,8 @@ func extractOpenAIRequestMetaFromBody(body []byte) (model string, stream bool, p
 }
 
 // normalizeOpenAIPassthroughOAuthBody 将透传 OAuth 请求体收敛为旧链路关键行为：
-// 1) store=false 2) stream=true
-func normalizeOpenAIPassthroughOAuthBody(body []byte) ([]byte, bool, error) {
+// 1) store=false 2) 非 compact 保持 stream=true；compact 强制 stream=false
+func normalizeOpenAIPassthroughOAuthBody(body []byte, compact bool) ([]byte, bool, error) {
 	if len(body) == 0 {
 		return body, false, nil
 	}
@@ -3748,22 +3969,40 @@ func normalizeOpenAIPassthroughOAuthBody(body []byte) ([]byte, bool, error) {
 	normalized := body
 	changed := false
 
-	if store := gjson.GetBytes(normalized, "store"); !store.Exists() || store.Type != gjson.False {
-		next, err := sjson.SetBytes(normalized, "store", false)
-		if err != nil {
-			return body, false, fmt.Errorf("normalize passthrough body store=false: %w", err)
+	if compact {
+		if store := gjson.GetBytes(normalized, "store"); store.Exists() {
+			next, err := sjson.DeleteBytes(normalized, "store")
+			if err != nil {
+				return body, false, fmt.Errorf("normalize passthrough body delete store: %w", err)
+			}
+			normalized = next
+			changed = true
 		}
-		normalized = next
-		changed = true
-	}
-
-	if stream := gjson.GetBytes(normalized, "stream"); !stream.Exists() || stream.Type != gjson.True {
-		next, err := sjson.SetBytes(normalized, "stream", true)
-		if err != nil {
-			return body, false, fmt.Errorf("normalize passthrough body stream=true: %w", err)
+		if stream := gjson.GetBytes(normalized, "stream"); stream.Exists() {
+			next, err := sjson.DeleteBytes(normalized, "stream")
+			if err != nil {
+				return body, false, fmt.Errorf("normalize passthrough body delete stream: %w", err)
+			}
+			normalized = next
+			changed = true
+		}
+	} else {
+		if store := gjson.GetBytes(normalized, "store"); !store.Exists() || store.Type != gjson.False {
+			next, err := sjson.SetBytes(normalized, "store", false)
+			if err != nil {
+				return body, false, fmt.Errorf("normalize passthrough body store=false: %w", err)
+			}
+			normalized = next
+			changed = true
+		}
+		if stream := gjson.GetBytes(normalized, "stream"); !stream.Exists() || stream.Type != gjson.True {
+			next, err := sjson.SetBytes(normalized, "stream", true)
+			if err != nil {
+				return body, false, fmt.Errorf("normalize passthrough body stream=true: %w", err)
+			}
+			normalized = next
+			changed = true
 		}
-		normalized = next
-		changed = true
 	}
 
 	return normalized, changed, nil

backend/internal/service/openai_gateway_service_test.go

@@ -14,6 +14,7 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/cespare/xxhash/v2"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
@@ -28,6 +29,22 @@ type stubOpenAIAccountRepo struct {
 	accounts []Account
 }
 
+type snapshotUpdateAccountRepo struct {
+	stubOpenAIAccountRepo
+	updateExtraCalls chan map[string]any
+}
+
+func (r *snapshotUpdateAccountRepo) UpdateExtra(ctx context.Context, id int64, updates map[string]any) error {
+	if r.updateExtraCalls != nil {
+		copied := make(map[string]any, len(updates))
+		for k, v := range updates {
+			copied[k] = v
+		}
+		r.updateExtraCalls <- copied
+	}
+	return nil
+}
+
 func (r stubOpenAIAccountRepo) GetByID(ctx context.Context, id int64) (*Account, error) {
 	for i := range r.accounts {
 		if r.accounts[i].ID == id {
@@ -1248,8 +1265,157 @@ func TestOpenAIValidateUpstreamBaseURLEnabledEnforcesAllowlist(t *testing.T) {
 	}
 }
 
+func TestOpenAIUpdateCodexUsageSnapshotFromHeaders(t *testing.T) {
+	repo := &snapshotUpdateAccountRepo{updateExtraCalls: make(chan map[string]any, 1)}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+	headers := http.Header{}
+	headers.Set("x-codex-primary-used-percent", "12")
+	headers.Set("x-codex-secondary-used-percent", "34")
+	headers.Set("x-codex-primary-window-minutes", "300")
+	headers.Set("x-codex-secondary-window-minutes", "10080")
+	headers.Set("x-codex-primary-reset-after-seconds", "600")
+	headers.Set("x-codex-secondary-reset-after-seconds", "86400")
+
+	svc.UpdateCodexUsageSnapshotFromHeaders(context.Background(), 123, headers)
+
+	select {
+	case updates := <-repo.updateExtraCalls:
+		require.Equal(t, 12.0, updates["codex_5h_used_percent"])
+		require.Equal(t, 34.0, updates["codex_7d_used_percent"])
+		require.Equal(t, 600, updates["codex_5h_reset_after_seconds"])
+		require.Equal(t, 86400, updates["codex_7d_reset_after_seconds"])
+	case <-time.After(2 * time.Second):
+		t.Fatal("expected UpdateExtra to be called")
+	}
+}
+
+func TestOpenAIResponsesRequestPathSuffix(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+
+	tests := []struct {
+		name string
+		path string
+		want string
+	}{
+		{name: "exact v1 responses", path: "/v1/responses", want: ""},
+		{name: "compact v1 responses", path: "/v1/responses/compact", want: "/compact"},
+		{name: "compact alias responses", path: "/responses/compact/", want: "/compact"},
+		{name: "nested suffix", path: "/openai/v1/responses/compact/detail", want: "/compact/detail"},
+		{name: "unrelated path", path: "/v1/chat/completions", want: ""},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			c.Request = httptest.NewRequest(http.MethodPost, tt.path, nil)
+			require.Equal(t, tt.want, openAIResponsesRequestPathSuffix(c))
+		})
+	}
+}
+
+func TestOpenAIBuildUpstreamRequestOpenAIPassthroughPreservesCompactPath(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses/compact", bytes.NewReader([]byte(`{"model":"gpt-5"}`)))
+
+	svc := &OpenAIGatewayService{}
+	account := &Account{Type: AccountTypeOAuth}
+
+	req, err := svc.buildUpstreamRequestOpenAIPassthrough(c.Request.Context(), c, account, []byte(`{"model":"gpt-5"}`), "token")
+	require.NoError(t, err)
+	require.Equal(t, chatgptCodexURL+"/compact", req.URL.String())
+	require.Equal(t, "application/json", req.Header.Get("Accept"))
+	require.Equal(t, codexCLIVersion, req.Header.Get("Version"))
+	require.NotEmpty(t, req.Header.Get("Session_Id"))
+}
+
+func TestOpenAIBuildUpstreamRequestCompactForcesJSONAcceptForOAuth(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses/compact", bytes.NewReader([]byte(`{"model":"gpt-5"}`)))
+
+	svc := &OpenAIGatewayService{}
+	account := &Account{
+		Type:        AccountTypeOAuth,
+		Credentials: map[string]any{"chatgpt_account_id": "chatgpt-acc"},
+	}
+
+	req, err := svc.buildUpstreamRequest(c.Request.Context(), c, account, []byte(`{"model":"gpt-5"}`), "token", false, "", true)
+	require.NoError(t, err)
+	require.Equal(t, chatgptCodexURL+"/compact", req.URL.String())
+	require.Equal(t, "application/json", req.Header.Get("Accept"))
+	require.Equal(t, codexCLIVersion, req.Header.Get("Version"))
+	require.NotEmpty(t, req.Header.Get("Session_Id"))
+}
+
+func TestOpenAIBuildUpstreamRequestPreservesCompactPathForAPIKeyBaseURL(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/responses/compact", bytes.NewReader([]byte(`{"model":"gpt-5"}`)))
+
+	svc := &OpenAIGatewayService{cfg: &config.Config{
+		Security: config.SecurityConfig{
+			URLAllowlist: config.URLAllowlistConfig{Enabled: false},
+		},
+	}}
+	account := &Account{
+		Type:        AccountTypeAPIKey,
+		Platform:    PlatformOpenAI,
+		Credentials: map[string]any{"base_url": "https://example.com/v1"},
+	}
+
+	req, err := svc.buildUpstreamRequest(c.Request.Context(), c, account, []byte(`{"model":"gpt-5"}`), "token", false, "", false)
+	require.NoError(t, err)
+	require.Equal(t, "https://example.com/v1/responses/compact", req.URL.String())
+}
+
+func TestOpenAIBuildUpstreamRequestOAuthOfficialClientOriginatorCompatibility(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name           string
+		userAgent      string
+		originator     string
+		wantOriginator string
+	}{
+		{name: "desktop originator preserved", originator: "Codex Desktop", wantOriginator: "Codex Desktop"},
+		{name: "vscode originator preserved", originator: "codex_vscode", wantOriginator: "codex_vscode"},
+		{name: "official ua fallback to codex_cli_rs", userAgent: "Codex Desktop/1.2.3", wantOriginator: "codex_cli_rs"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader([]byte(`{"model":"gpt-5"}`)))
+			if tt.userAgent != "" {
+				c.Request.Header.Set("User-Agent", tt.userAgent)
+			}
+			if tt.originator != "" {
+				c.Request.Header.Set("originator", tt.originator)
+			}
+
+			svc := &OpenAIGatewayService{}
+			account := &Account{
+				Type:        AccountTypeOAuth,
+				Credentials: map[string]any{"chatgpt_account_id": "chatgpt-acc"},
+			}
+
+			isCodexCLI := openai.IsCodexOfficialClientByHeaders(c.GetHeader("User-Agent"), c.GetHeader("originator"))
+			req, err := svc.buildUpstreamRequest(c.Request.Context(), c, account, []byte(`{"model":"gpt-5"}`), "token", false, "", isCodexCLI)
+			require.NoError(t, err)
+			require.Equal(t, tt.wantOriginator, req.Header.Get("originator"))
+		})
+	}
+}
+
 // ==================== P1-08 修复：model 替换性能优化测试 ====================
 
+// ==================== P1-08 修复：model 替换性能优化测试 =============
 func TestReplaceModelInSSELine(t *testing.T) {
 	svc := &OpenAIGatewayService{}
 
@@ -1576,3 +1742,27 @@ func TestHandleOAuthSSEToJSON_NoFinalResponseKeepsSSEBody(t *testing.T) {
 	require.Contains(t, rec.Header().Get("Content-Type"), "text/event-stream")
 	require.Contains(t, rec.Body.String(), `data: {"type":"response.in_progress"`)
 }
+
+func TestHandleOAuthSSEToJSON_ResponseFailedReturnsProtocolError(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	svc := &OpenAIGatewayService{cfg: &config.Config{}}
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+	}
+	body := []byte(strings.Join([]string{
+		`data: {"type":"response.failed","error":{"message":"upstream rejected request"}}`,
+		`data: [DONE]`,
+	}, "\n"))
+
+	usage, err := svc.handleOAuthSSEToJSON(resp, c, body, "gpt-4o", "gpt-4o")
+	require.Nil(t, usage)
+	require.Error(t, err)
+	require.Equal(t, http.StatusBadGateway, rec.Code)
+	require.Contains(t, rec.Body.String(), "upstream rejected request")
+	require.Contains(t, rec.Header().Get("Content-Type"), "application/json")
+}

backend/internal/service/openai_oauth_passthrough_test.go

@@ -236,6 +236,60 @@ func TestOpenAIGatewayService_OAuthPassthrough_StreamKeepsToolNameAndBodyNormali
 	require.NotContains(t, body, "\"name\":\"edit\"")
 }
 
+func TestOpenAIGatewayService_OAuthPassthrough_CompactUsesJSONAndKeepsNonStreaming(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses/compact", bytes.NewReader(nil))
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.1.0")
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	originalBody := []byte(`{"model":"gpt-5.1-codex","stream":true,"store":true,"instructions":"local-test-instructions","input":[{"type":"text","text":"compact me"}]}`)
+
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}, "x-request-id": []string{"rid-compact"}},
+		Body:       io.NopCloser(strings.NewReader(`{"id":"cmp_123","usage":{"input_tokens":11,"output_tokens":22}}`)),
+	}
+	upstream := &httpUpstreamRecorder{resp: resp}
+
+	svc := &OpenAIGatewayService{
+		cfg:          &config.Config{Gateway: config.GatewayConfig{ForceCodexCLI: false}},
+		httpUpstream: upstream,
+	}
+
+	account := &Account{
+		ID:             123,
+		Name:           "acc",
+		Platform:       PlatformOpenAI,
+		Type:           AccountTypeOAuth,
+		Concurrency:    1,
+		Credentials:    map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"},
+		Extra:          map[string]any{"openai_passthrough": true},
+		Status:         StatusActive,
+		Schedulable:    true,
+		RateMultiplier: f64p(1),
+	}
+
+	result, err := svc.Forward(context.Background(), c, account, originalBody)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.False(t, result.Stream)
+
+	require.False(t, gjson.GetBytes(upstream.lastBody, "store").Exists())
+	require.False(t, gjson.GetBytes(upstream.lastBody, "stream").Exists())
+	require.Equal(t, "gpt-5.1-codex", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.Equal(t, "compact me", gjson.GetBytes(upstream.lastBody, "input.0.text").String())
+	require.Equal(t, "local-test-instructions", strings.TrimSpace(gjson.GetBytes(upstream.lastBody, "instructions").String()))
+	require.Equal(t, "application/json", upstream.lastReq.Header.Get("Accept"))
+	require.Equal(t, codexCLIVersion, upstream.lastReq.Header.Get("Version"))
+	require.NotEmpty(t, upstream.lastReq.Header.Get("Session_Id"))
+	require.Equal(t, "chatgpt.com", upstream.lastReq.Host)
+	require.Equal(t, "chatgpt-acc", upstream.lastReq.Header.Get("chatgpt-account-id"))
+	require.Contains(t, rec.Body.String(), `"id":"cmp_123"`)
+}
+
 func TestOpenAIGatewayService_OAuthPassthrough_CodexMissingInstructionsRejectedBeforeUpstream(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	logSink, restore := captureStructuredLog(t)

backend/internal/service/openai_ws_forwarder.go

@@ -1141,11 +1141,7 @@ func (s *OpenAIGatewayService) buildOpenAIWSHeaders(
 		if chatgptAccountID := account.GetChatGPTAccountID(); chatgptAccountID != "" {
 			headers.Set("chatgpt-account-id", chatgptAccountID)
 		}
-		if isCodexCLI {
-			headers.Set("originator", "codex_cli_rs")
-		} else {
-			headers.Set("originator", "opencode")
-		}
+		headers.Set("originator", resolveOpenAIUpstreamOriginator(c, isCodexCLI))
 	}
 
 	betaValue := openAIWSBetaV2Value
@@ -2309,6 +2305,7 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 		ReasoningEffort: extractOpenAIReasoningEffort(reqBody, originalModel),
 		Stream:          reqStream,
 		OpenAIWSMode:    true,
+		ResponseHeaders: lease.HandshakeHeaders(),
 		Duration:        time.Since(startTime),
 		FirstTokenMs:    firstTokenMs,
 	}, nil
@@ -2542,7 +2539,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 		}
 	}
 
-	isCodexCLI := openai.IsCodexCLIRequest(c.GetHeader("User-Agent")) || (s.cfg != nil && s.cfg.Gateway.ForceCodexCLI)
+	isCodexCLI := openai.IsCodexOfficialClientByHeaders(c.GetHeader("User-Agent"), c.GetHeader("originator")) || (s.cfg != nil && s.cfg.Gateway.ForceCodexCLI)
 	wsHeaders, _ := s.buildOpenAIWSHeaders(c, account, token, wsDecision, isCodexCLI, turnState, strings.TrimSpace(c.GetHeader(openAIWSTurnMetadataHeader)), firstPayload.promptCacheKey)
 	baseAcquireReq := openAIWSAcquireRequest{
 		Account: account,
@@ -2919,6 +2916,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 					ReasoningEffort: extractOpenAIReasoningEffortFromBody(payload, originalModel),
 					Stream:          reqStream,
 					OpenAIWSMode:    true,
+					ResponseHeaders: lease.HandshakeHeaders(),
 					Duration:        time.Since(turnStart),
 					FirstTokenMs:    firstTokenMs,
 				}, nil

backend/internal/service/openai_ws_forwarder_success_test.go

@@ -458,6 +458,86 @@ func TestOpenAIGatewayService_Forward_WSv2_OAuthStoreFalseByDefault(t *testing.T
 	require.Equal(t, "conv-oauth-1", captureDialer.lastHeaders.Get("conversation_id"))
 }
 
+func TestOpenAIGatewayService_Forward_WSv2_OAuthOriginatorCompatibility(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name           string
+		userAgent      string
+		originator     string
+		wantOriginator string
+	}{
+		{name: "desktop originator preserved", originator: "Codex Desktop", wantOriginator: "Codex Desktop"},
+		{name: "vscode originator preserved", originator: "codex_vscode", wantOriginator: "codex_vscode"},
+		{name: "official ua fallback to codex_cli_rs", userAgent: "Codex Desktop/1.2.3", wantOriginator: "codex_cli_rs"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+			if tt.userAgent != "" {
+				c.Request.Header.Set("User-Agent", tt.userAgent)
+			}
+			if tt.originator != "" {
+				c.Request.Header.Set("originator", tt.originator)
+			}
+
+			cfg := &config.Config{}
+			cfg.Security.URLAllowlist.Enabled = false
+			cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+			cfg.Gateway.OpenAIWS.Enabled = true
+			cfg.Gateway.OpenAIWS.OAuthEnabled = true
+			cfg.Gateway.OpenAIWS.APIKeyEnabled = true
+			cfg.Gateway.OpenAIWS.ResponsesWebsocketsV2 = true
+			cfg.Gateway.OpenAIWS.AllowStoreRecovery = false
+			cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
+			cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+			cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
+
+			captureConn := &openAIWSCaptureConn{
+				events: [][]byte{
+					[]byte(`{"type":"response.completed","response":{"id":"resp_oauth_originator","model":"gpt-5.1","usage":{"input_tokens":1,"output_tokens":1}}}`),
+				},
+			}
+			captureDialer := &openAIWSCaptureDialer{conn: captureConn}
+			pool := newOpenAIWSConnPool(cfg)
+			pool.setClientDialerForTest(captureDialer)
+
+			svc := &OpenAIGatewayService{
+				cfg:              cfg,
+				httpUpstream:     &httpUpstreamRecorder{},
+				cache:            &stubGatewayCache{},
+				openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+				toolCorrector:    NewCodexToolCorrector(),
+				openaiWSPool:     pool,
+			}
+			account := &Account{
+				ID:          129,
+				Name:        "openai-oauth",
+				Platform:    PlatformOpenAI,
+				Type:        AccountTypeOAuth,
+				Status:      StatusActive,
+				Schedulable: true,
+				Concurrency: 1,
+				Credentials: map[string]any{
+					"access_token": "oauth-token-1",
+				},
+				Extra: map[string]any{
+					"responses_websockets_v2_enabled": true,
+				},
+			}
+
+			body := []byte(`{"model":"gpt-5.1","stream":false,"input":[{"type":"input_text","text":"hello"}]}`)
+			result, err := svc.Forward(context.Background(), c, account, body)
+			require.NoError(t, err)
+			require.NotNil(t, result)
+			require.Equal(t, tt.wantOriginator, captureDialer.lastHeaders.Get("originator"))
+		})
+	}
+}
+
 func TestOpenAIGatewayService_Forward_WSv2_HeaderSessionFallbackFromPromptCacheKey(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 

backend/internal/service/openai_ws_pool.go

@@ -126,6 +126,13 @@ func (l *openAIWSConnLease) HandshakeHeader(name string) string {
 	return l.conn.handshakeHeader(name)
 }
 
+func (l *openAIWSConnLease) HandshakeHeaders() http.Header {
+	if l == nil || l.conn == nil {
+		return nil
+	}
+	return cloneHeader(l.conn.handshakeHeaders)
+}
+
 func (l *openAIWSConnLease) IsPrewarmed() bool {
 	if l == nil || l.conn == nil {
 		return false

backend/internal/service/openai_ws_protocol_forward_test.go

@@ -391,6 +391,7 @@ func TestNewOpenAIGatewayService_InitializesOpenAIWSResolver(t *testing.T) {
 		nil,
 		nil,
 		nil,
+		nil,
 		cfg,
 		nil,
 		nil,

backend/internal/service/openai_ws_v2_passthrough_adapter.go

@@ -107,7 +107,7 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 
 	isCodexCLI := false
 	if c != nil {
-		isCodexCLI = openai.IsCodexCLIRequest(c.GetHeader("User-Agent"))
+		isCodexCLI = openai.IsCodexOfficialClientByHeaders(c.GetHeader("User-Agent"), c.GetHeader("originator"))
 	}
 	if s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
 		isCodexCLI = true
@@ -177,11 +177,12 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 						CacheCreationInputTokens: turn.Usage.CacheCreationInputTokens,
 						CacheReadInputTokens:     turn.Usage.CacheReadInputTokens,
 					},
-					Model:        turn.RequestModel,
-					Stream:       true,
-					OpenAIWSMode: true,
-					Duration:     turn.Duration,
-					FirstTokenMs: turn.FirstTokenMs,
+					Model:           turn.RequestModel,
+					Stream:          true,
+					OpenAIWSMode:    true,
+					ResponseHeaders: cloneHeader(handshakeHeaders),
+					Duration:        turn.Duration,
+					FirstTokenMs:    turn.FirstTokenMs,
 				}
 				logOpenAIWSV2Passthrough(
 					"relay_turn_completed account_id=%d turn=%d request_id=%s terminal_event=%s duration_ms=%d first_token_ms=%d input_tokens=%d output_tokens=%d cache_read_tokens=%d",
@@ -223,11 +224,12 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 			CacheCreationInputTokens: relayResult.Usage.CacheCreationInputTokens,
 			CacheReadInputTokens:     relayResult.Usage.CacheReadInputTokens,
 		},
-		Model:        relayResult.RequestModel,
-		Stream:       true,
-		OpenAIWSMode: true,
-		Duration:     relayResult.Duration,
-		FirstTokenMs: relayResult.FirstTokenMs,
+		Model:           relayResult.RequestModel,
+		Stream:          true,
+		OpenAIWSMode:    true,
+		ResponseHeaders: cloneHeader(handshakeHeaders),
+		Duration:        relayResult.Duration,
+		FirstTokenMs:    relayResult.FirstTokenMs,
 	}
 
 	turnCount := int(completedTurns.Load())

backend/internal/service/prompts/codex_cli_instructions.md

@@ -1,275 +0,0 @@
-You are a coding agent running in the Codex CLI, a terminal-based coding assistant. Codex CLI is an open source project led by OpenAI. You are expected to be precise, safe, and helpful.
-
-Your capabilities:
-
-- Receive user prompts and other context provided by the harness, such as files in the workspace.
-- Communicate with the user by streaming thinking & responses, and by making & updating plans.
-- Emit function calls to run terminal commands and apply patches. Depending on how this specific run is configured, you can request that these function calls be escalated to the user for approval before running. More on this in the "Sandbox and approvals" section.
-
-Within this context, Codex refers to the open-source agentic coding interface (not the old Codex language model built by OpenAI).
-
-# How you work
-
-## Personality
-
-Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
-
-# AGENTS.md spec
-- Repos often contain AGENTS.md files. These files can appear anywhere within the repository.
-- These files are a way for humans to give you (the agent) instructions or tips for working within the container.
-- Some examples might be: coding conventions, info about how code is organized, or instructions for how to run or test code.
-- Instructions in AGENTS.md files:
-    - The scope of an AGENTS.md file is the entire directory tree rooted at the folder that contains it.
-    - For every file you touch in the final patch, you must obey instructions in any AGENTS.md file whose scope includes that file.
-    - Instructions about code style, structure, naming, etc. apply only to code within the AGENTS.md file's scope, unless the file states otherwise.
-    - More-deeply-nested AGENTS.md files take precedence in the case of conflicting instructions.
-    - Direct system/developer/user instructions (as part of a prompt) take precedence over AGENTS.md instructions.
-- The contents of the AGENTS.md file at the root of the repo and any directories from the CWD up to the root are included with the developer message and don't need to be re-read. When working in a subdirectory of CWD, or a directory outside the CWD, check for any AGENTS.md files that may be applicable.
-
-## Responsiveness
-
-### Preamble messages
-
-Before making tool calls, send a brief preamble to the user explaining what you’re about to do. When sending preamble messages, follow these principles and examples:
-
-- **Logically group related actions**: if you’re about to run several related commands, describe them together in one preamble rather than sending a separate note for each.
-- **Keep it concise**: be no more than 1-2 sentences, focused on immediate, tangible next steps. (8–12 words for quick updates).
-- **Build on prior context**: if this is not your first tool call, use the preamble message to connect the dots with what’s been done so far and create a sense of momentum and clarity for the user to understand your next actions.
-- **Keep your tone light, friendly and curious**: add small touches of personality in preambles feel collaborative and engaging.
-- **Exception**: Avoid adding a preamble for every trivial read (e.g., `cat` a single file) unless it’s part of a larger grouped action.
-
-**Examples:**
-
-- “I’ve explored the repo; now checking the API route definitions.”
-- “Next, I’ll patch the config and update the related tests.”
-- “I’m about to scaffold the CLI commands and helper functions.”
-- “Ok cool, so I’ve wrapped my head around the repo. Now digging into the API routes.”
-- “Config’s looking tidy. Next up is patching helpers to keep things in sync.”
-- “Finished poking at the DB gateway. I will now chase down error handling.”
-- “Alright, build pipeline order is interesting. Checking how it reports failures.”
-- “Spotted a clever caching util; now hunting where it gets used.”
-
-## Planning
-
-You have access to an `update_plan` tool which tracks steps and progress and renders them to the user. Using the tool helps demonstrate that you've understood the task and convey how you're approaching it. Plans can help to make complex, ambiguous, or multi-phase work clearer and more collaborative for the user. A good plan should break the task into meaningful, logically ordered steps that are easy to verify as you go.
-
-Note that plans are not for padding out simple work with filler steps or stating the obvious. The content of your plan should not involve doing anything that you aren't capable of doing (i.e. don't try to test things that you can't test). Do not use plans for simple or single-step queries that you can just do or answer immediately.
-
-Do not repeat the full contents of the plan after an `update_plan` call — the harness already displays it. Instead, summarize the change made and highlight any important context or next step.
-
-Before running a command, consider whether or not you have completed the previous step, and make sure to mark it as completed before moving on to the next step. It may be the case that you complete all steps in your plan after a single pass of implementation. If this is the case, you can simply mark all the planned steps as completed. Sometimes, you may need to change plans in the middle of a task: call `update_plan` with the updated plan and make sure to provide an `explanation` of the rationale when doing so.
-
-Use a plan when:
-
-- The task is non-trivial and will require multiple actions over a long time horizon.
-- There are logical phases or dependencies where sequencing matters.
-- The work has ambiguity that benefits from outlining high-level goals.
-- You want intermediate checkpoints for feedback and validation.
-- When the user asked you to do more than one thing in a single prompt
-- The user has asked you to use the plan tool (aka "TODOs")
-- You generate additional steps while working, and plan to do them before yielding to the user
-
-### Examples
-
-**High-quality plans**
-
-Example 1:
-
-1. Add CLI entry with file args
-2. Parse Markdown via CommonMark library
-3. Apply semantic HTML template
-4. Handle code blocks, images, links
-5. Add error handling for invalid files
-
-Example 2:
-
-1. Define CSS variables for colors
-2. Add toggle with localStorage state
-3. Refactor components to use variables
-4. Verify all views for readability
-5. Add smooth theme-change transition
-
-Example 3:
-
-1. Set up Node.js + WebSocket server
-2. Add join/leave broadcast events
-3. Implement messaging with timestamps
-4. Add usernames + mention highlighting
-5. Persist messages in lightweight DB
-6. Add typing indicators + unread count
-
-**Low-quality plans**
-
-Example 1:
-
-1. Create CLI tool
-2. Add Markdown parser
-3. Convert to HTML
-
-Example 2:
-
-1. Add dark mode toggle
-2. Save preference
-3. Make styles look good
-
-Example 3:
-
-1. Create single-file HTML game
-2. Run quick sanity check
-3. Summarize usage instructions
-
-If you need to write a plan, only write high quality plans, not low quality ones.
-
-## Task execution
-
-You are a coding agent. Please keep going until the query is completely resolved, before ending your turn and yielding back to the user. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability, using the tools available to you, before coming back to the user. Do NOT guess or make up an answer.
-
-You MUST adhere to the following criteria when solving queries:
-
-- Working on the repo(s) in the current environment is allowed, even if they are proprietary.
-- Analyzing code for vulnerabilities is allowed.
-- Showing user code and tool call details is allowed.
-- Use the `apply_patch` tool to edit files (NEVER try `applypatch` or `apply-patch`, only `apply_patch`): {"command":["apply_patch","*** Begin Patch\\n*** Update File: path/to/file.py\\n@@ def example():\\n- pass\\n+ return 123\\n*** End Patch"]}
-
-If completing the user's task requires writing or modifying files, your code and final answer should follow these coding guidelines, though user instructions (i.e. AGENTS.md) may override these guidelines:
-
-- Fix the problem at the root cause rather than applying surface-level patches, when possible.
-- Avoid unneeded complexity in your solution.
-- Do not attempt to fix unrelated bugs or broken tests. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
-- Update documentation as necessary.
-- Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task.
-- Use `git log` and `git blame` to search the history of the codebase if additional context is required.
-- NEVER add copyright or license headers unless specifically requested.
-- Do not waste tokens by re-reading files after calling `apply_patch` on them. The tool call will fail if it didn't work. The same goes for making folders, deleting folders, etc.
-- Do not `git commit` your changes or create new git branches unless explicitly requested.
-- Do not add inline comments within code unless explicitly requested.
-- Do not use one-letter variable names unless explicitly requested.
-- NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
-
-## Validating your work
-
-If the codebase has tests or the ability to build or run, consider using them to verify that your work is complete. 
-
-When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
-
-Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
-
-For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
-
-Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
-
-- When running in non-interactive approval modes like **never** or **on-failure**, proactively run tests, lint and do whatever you need to ensure you've completed the task.
-- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
-- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
-
-## Ambition vs. precision
-
-For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.
-
-If you're operating in an existing codebase, you should make sure you do exactly what the user asks with surgical precision. Treat the surrounding codebase with respect, and don't overstep (i.e. changing filenames or variables unnecessarily). You should balance being sufficiently ambitious and proactive when completing tasks of this nature.
-
-You should use judicious initiative to decide on the right level of detail and complexity to deliver based on the user's needs. This means showing good judgment that you're capable of doing the right extras without gold-plating. This might be demonstrated by high-value, creative touches when scope of the task is vague; while being surgical and targeted when scope is tightly specified.
-
-## Sharing progress updates
-
-For especially longer tasks that you work on (i.e. requiring many tool calls, or a plan with multiple steps), you should provide progress updates back to the user at reasonable intervals. These updates should be structured as a concise sentence or two (no more than 8-10 words long) recapping progress so far in plain language: this update demonstrates your understanding of what needs to be done, progress so far (i.e. files explores, subtasks complete), and where you're going next.
-
-Before doing large chunks of work that may incur latency as experienced by the user (i.e. writing a new file), you should send a concise message to the user with an update indicating what you're about to do to ensure they know what you're spending time on. Don't start editing or writing large files before informing the user what you are doing and why.
-
-The messages you send before tool calls should describe what is immediately about to be done next in very concise language. If there was previous work done, this preamble message should also include a note about the work done so far to bring the user along.
-
-## Presenting your work and final message
-
-Your final message should read naturally, like an update from a concise teammate. For casual conversation, brainstorming tasks, or quick questions from the user, respond in a friendly, conversational tone. You should ask questions, suggest ideas, and adapt to the user’s style. If you've finished a large amount of work, when describing what you've done to the user, you should follow the final answer formatting guidelines to communicate substantive changes. You don't need to add structured formatting for one-word answers, greetings, or purely conversational exchanges.
-
-You can skip heavy formatting for single, simple actions or confirmations. In these cases, respond in plain sentences with any relevant next step or quick option. Reserve multi-section structured responses for results that need grouping or explanation.
-
-The user is working on the same computer as you, and has access to your work. As such there's no need to show the full contents of large files you have already written unless the user explicitly asks for them. Similarly, if you've created or modified files using `apply_patch`, there's no need to tell users to "save the file" or "copy the code into a file"—just reference the file path.
-
-If there's something that you think you could help with as a logical next step, concisely ask the user if they want you to do so. Good examples of this are running tests, committing changes, or building out the next logical component. If there’s something that you couldn't do (even with approval) but that the user might want to do (such as verifying changes by running the app), include those instructions succinctly.
-
-Brevity is very important as a default. You should be very concise (i.e. no more than 10 lines), but can relax this requirement for tasks where additional detail and comprehensiveness is important for the user's understanding.
-
-### Final answer structure and style guidelines
-
-You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
-
-**Section Headers**
-
-- Use only when they improve clarity — they are not mandatory for every answer.
-- Choose descriptive names that fit the content
-- Keep headers short (1–3 words) and in `**Title Case**`. Always start headers with `**` and end with `**`
-- Leave no blank line before the first bullet under a header.
-- Section headers should only be used where they genuinely improve scanability; avoid fragmenting the answer.
-
-**Bullets**
-
-- Use `-` followed by a space for every bullet.
-- Merge related points when possible; avoid a bullet for every trivial detail.
-- Keep bullets to one line unless breaking for clarity is unavoidable.
-- Group into short lists (4–6 bullets) ordered by importance.
-- Use consistent keyword phrasing and formatting across sections.
-
-**Monospace**
-
-- Wrap all commands, file paths, env vars, and code identifiers in backticks (`` `...` ``).
-- Apply to inline examples and to bullet keywords if the keyword itself is a literal file/command.
-- Never mix monospace and bold markers; choose one based on whether it’s a keyword (`**`) or inline code/path (`` ` ``).
-
-**File References**
-When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
-  * Use inline code to make file paths clickable.
-  * Each reference should have a stand alone path. Even if it's the same file.
-  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
-  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
-  * Do not use URIs like file://, vscode://, or https://.
-  * Do not provide range of lines
-  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
-
-**Structure**
-
-- Place related bullets together; don’t mix unrelated concepts in the same section.
-- Order sections from general → specific → supporting info.
-- For subsections (e.g., “Binaries” under “Rust Workspace”), introduce with a bolded keyword bullet, then list items under it.
-- Match structure to complexity:
-  - Multi-part or detailed results → use clear headers and grouped bullets.
-  - Simple results → minimal headers, possibly just a short list or paragraph.
-
-**Tone**
-
-- Keep the voice collaborative and natural, like a coding partner handing off work.
-- Be concise and factual — no filler or conversational commentary and avoid unnecessary repetition
-- Use present tense and active voice (e.g., “Runs tests” not “This will run tests”).
-- Keep descriptions self-contained; don’t refer to “above” or “below”.
-- Use parallel structure in lists for consistency.
-
-**Don’t**
-
-- Don’t use literal words “bold” or “monospace” in the content.
-- Don’t nest bullets or create deep hierarchies.
-- Don’t output ANSI escape codes directly — the CLI renderer applies them.
-- Don’t cram unrelated keywords into a single bullet; split for clarity.
-- Don’t let keyword lists run long — wrap or reformat for scanability.
-
-Generally, ensure your final answers adapt their shape and depth to the request. For example, answers to code explanations should have a precise, structured explanation with code references that answer the question directly. For tasks with a simple implementation, lead with the outcome and supplement only with what’s needed for clarity. Larger changes can be presented as a logical walkthrough of your approach, grouping related steps, explaining rationale where it adds value, and highlighting next actions to accelerate the user. Your answers should provide the right level of detail while being easily scannable.
-
-For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.
-
-# Tool Guidelines
-
-## Shell commands
-
-When using the shell, you must adhere to the following guidelines:
-
-- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
-- Do not use python scripts to attempt to output larger chunks of a file.
-
-## `update_plan`
-
-A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.
-
-To create a new plan, call `update_plan` with a short list of 1‑sentence steps (no more than 5-7 words each) with a `status` for each step (`pending`, `in_progress`, or `completed`).
-
-When steps have been completed, use `update_plan` to mark each finished step as `completed` and the next step you are working on as `in_progress`. There should always be exactly one `in_progress` step until everything is done. You can mark multiple items as complete in a single `update_plan` call.
-
-If all steps are complete, ensure you call `update_plan` to mark all steps as `completed`.

backend/internal/service/ratelimit_service.go

@@ -970,12 +970,27 @@ func (s *RateLimitService) UpdateSessionWindow(ctx context.Context, account *Acc
 		windowStart = &start
 		windowEnd = &end
 		slog.Info("account_session_window_initialized", "account_id", account.ID, "window_start", start, "window_end", end, "status", status)
+		// 窗口重置时清除旧的 utilization，避免残留上个窗口的数据
+		_ = s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+			"session_window_utilization": nil,
+		})
 	}
 
 	if err := s.accountRepo.UpdateSessionWindow(ctx, account.ID, windowStart, windowEnd, status); err != nil {
 		slog.Warn("session_window_update_failed", "account_id", account.ID, "error", err)
 	}
 
+	// 存储真实的 utilization 值（0-1 小数），供 estimateSetupTokenUsage 使用
+	if utilStr := headers.Get("anthropic-ratelimit-unified-5h-utilization"); utilStr != "" {
+		if util, err := strconv.ParseFloat(utilStr, 64); err == nil {
+			if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+				"session_window_utilization": util,
+			}); err != nil {
+				slog.Warn("session_window_utilization_update_failed", "account_id", account.ID, "error", err)
+			}
+		}
+	}
+
 	// 如果状态为allowed且之前有限流，说明窗口已重置，清除限流状态
 	if status == "allowed" && account.IsRateLimited() {
 		if err := s.ClearRateLimit(ctx, account.ID); err != nil {

backend/internal/service/user_group_rate_resolver.go

@@ -0,0 +1,103 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	gocache "github.com/patrickmn/go-cache"
+	"golang.org/x/sync/singleflight"
+)
+
+type userGroupRateResolver struct {
+	repo         UserGroupRateRepository
+	cache        *gocache.Cache
+	cacheTTL     time.Duration
+	sf           *singleflight.Group
+	logComponent string
+}
+
+func newUserGroupRateResolver(repo UserGroupRateRepository, cache *gocache.Cache, cacheTTL time.Duration, sf *singleflight.Group, logComponent string) *userGroupRateResolver {
+	if cacheTTL <= 0 {
+		cacheTTL = defaultUserGroupRateCacheTTL
+	}
+	if cache == nil {
+		cache = gocache.New(cacheTTL, time.Minute)
+	}
+	if logComponent == "" {
+		logComponent = "service.gateway"
+	}
+	if sf == nil {
+		sf = &singleflight.Group{}
+	}
+
+	return &userGroupRateResolver{
+		repo:         repo,
+		cache:        cache,
+		cacheTTL:     cacheTTL,
+		sf:           sf,
+		logComponent: logComponent,
+	}
+}
+
+func (r *userGroupRateResolver) Resolve(ctx context.Context, userID, groupID int64, groupDefaultMultiplier float64) float64 {
+	if r == nil || userID <= 0 || groupID <= 0 {
+		return groupDefaultMultiplier
+	}
+
+	key := fmt.Sprintf("%d:%d", userID, groupID)
+	if r.cache != nil {
+		if cached, ok := r.cache.Get(key); ok {
+			if multiplier, castOK := cached.(float64); castOK {
+				userGroupRateCacheHitTotal.Add(1)
+				return multiplier
+			}
+		}
+	}
+	if r.repo == nil {
+		return groupDefaultMultiplier
+	}
+	userGroupRateCacheMissTotal.Add(1)
+
+	value, err, shared := r.sf.Do(key, func() (any, error) {
+		if r.cache != nil {
+			if cached, ok := r.cache.Get(key); ok {
+				if multiplier, castOK := cached.(float64); castOK {
+					userGroupRateCacheHitTotal.Add(1)
+					return multiplier, nil
+				}
+			}
+		}
+
+		userGroupRateCacheLoadTotal.Add(1)
+		userRate, repoErr := r.repo.GetByUserAndGroup(ctx, userID, groupID)
+		if repoErr != nil {
+			return nil, repoErr
+		}
+
+		multiplier := groupDefaultMultiplier
+		if userRate != nil {
+			multiplier = *userRate
+		}
+		if r.cache != nil {
+			r.cache.Set(key, multiplier, r.cacheTTL)
+		}
+		return multiplier, nil
+	})
+	if shared {
+		userGroupRateCacheSFSharedTotal.Add(1)
+	}
+	if err != nil {
+		userGroupRateCacheFallbackTotal.Add(1)
+		logger.LegacyPrintf(r.logComponent, "get user group rate failed, fallback to group default: user=%d group=%d err=%v", userID, groupID, err)
+		return groupDefaultMultiplier
+	}
+
+	multiplier, ok := value.(float64)
+	if !ok {
+		userGroupRateCacheFallbackTotal.Add(1)
+		return groupDefaultMultiplier
+	}
+	return multiplier
+}

backend/internal/service/user_group_rate_resolver_test.go

@@ -0,0 +1,83 @@
+package service
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	gocache "github.com/patrickmn/go-cache"
+	"github.com/stretchr/testify/require"
+)
+
+type userGroupRateResolverRepoStub struct {
+	UserGroupRateRepository
+
+	rate  *float64
+	err   error
+	calls int
+}
+
+func (s *userGroupRateResolverRepoStub) GetByUserAndGroup(ctx context.Context, userID, groupID int64) (*float64, error) {
+	s.calls++
+	if s.err != nil {
+		return nil, s.err
+	}
+	return s.rate, nil
+}
+
+func TestNewUserGroupRateResolver_Defaults(t *testing.T) {
+	resolver := newUserGroupRateResolver(nil, nil, 0, nil, "")
+
+	require.NotNil(t, resolver)
+	require.NotNil(t, resolver.cache)
+	require.Equal(t, defaultUserGroupRateCacheTTL, resolver.cacheTTL)
+	require.NotNil(t, resolver.sf)
+	require.Equal(t, "service.gateway", resolver.logComponent)
+}
+
+func TestUserGroupRateResolverResolve_FallbackForNilResolverAndInvalidIDs(t *testing.T) {
+	var nilResolver *userGroupRateResolver
+	require.Equal(t, 1.4, nilResolver.Resolve(context.Background(), 101, 202, 1.4))
+
+	resolver := newUserGroupRateResolver(nil, nil, time.Second, nil, "service.test")
+	require.Equal(t, 1.4, resolver.Resolve(context.Background(), 0, 202, 1.4))
+	require.Equal(t, 1.4, resolver.Resolve(context.Background(), 101, 0, 1.4))
+}
+
+func TestUserGroupRateResolverResolve_InvalidCacheEntryLoadsRepoAndCaches(t *testing.T) {
+	resetGatewayHotpathStatsForTest()
+
+	rate := 1.7
+	repo := &userGroupRateResolverRepoStub{rate: &rate}
+	cache := gocache.New(time.Minute, time.Minute)
+	cache.Set("101:202", "bad-cache", time.Minute)
+	resolver := newUserGroupRateResolver(repo, cache, time.Minute, nil, "service.test")
+
+	got := resolver.Resolve(context.Background(), 101, 202, 1.2)
+	require.Equal(t, rate, got)
+	require.Equal(t, 1, repo.calls)
+
+	cached, ok := cache.Get("101:202")
+	require.True(t, ok)
+	require.Equal(t, rate, cached)
+
+	hit, miss, load, _, fallback := GatewayUserGroupRateCacheStats()
+	require.Equal(t, int64(0), hit)
+	require.Equal(t, int64(1), miss)
+	require.Equal(t, int64(1), load)
+	require.Equal(t, int64(0), fallback)
+}
+
+func TestGatewayServiceGetUserGroupRateMultiplier_FallbacksAndUsesExistingResolver(t *testing.T) {
+	var nilSvc *GatewayService
+	require.Equal(t, 1.3, nilSvc.getUserGroupRateMultiplier(context.Background(), 101, 202, 1.3))
+
+	rate := 1.9
+	repo := &userGroupRateResolverRepoStub{rate: &rate}
+	resolver := newUserGroupRateResolver(repo, nil, time.Minute, nil, "service.gateway")
+	svc := &GatewayService{userGroupRateResolver: resolver}
+
+	got := svc.getUserGroupRateMultiplier(context.Background(), 101, 202, 1.2)
+	require.Equal(t, rate, got)
+	require.Equal(t, 1, repo.calls)
+}

backend/internal/web/embed_on.go

@@ -83,14 +83,7 @@ func (s *FrontendServer) Middleware() gin.HandlerFunc {
 		path := c.Request.URL.Path
 
 		// Skip API routes
-		if strings.HasPrefix(path, "/api/") ||
-			strings.HasPrefix(path, "/v1/") ||
-			strings.HasPrefix(path, "/v1beta/") ||
-			strings.HasPrefix(path, "/sora/") ||
-			strings.HasPrefix(path, "/antigravity/") ||
-			strings.HasPrefix(path, "/setup/") ||
-			path == "/health" ||
-			path == "/responses" {
+		if shouldBypassEmbeddedFrontend(path) {
 			c.Next()
 			return
 		}
@@ -207,14 +200,7 @@ func ServeEmbeddedFrontend() gin.HandlerFunc {
 	return func(c *gin.Context) {
 		path := c.Request.URL.Path
 
-		if strings.HasPrefix(path, "/api/") ||
-			strings.HasPrefix(path, "/v1/") ||
-			strings.HasPrefix(path, "/v1beta/") ||
-			strings.HasPrefix(path, "/sora/") ||
-			strings.HasPrefix(path, "/antigravity/") ||
-			strings.HasPrefix(path, "/setup/") ||
-			path == "/health" ||
-			path == "/responses" {
+		if shouldBypassEmbeddedFrontend(path) {
 			c.Next()
 			return
 		}
@@ -235,6 +221,19 @@ func ServeEmbeddedFrontend() gin.HandlerFunc {
 	}
 }
 
+func shouldBypassEmbeddedFrontend(path string) bool {
+	trimmed := strings.TrimSpace(path)
+	return strings.HasPrefix(trimmed, "/api/") ||
+		strings.HasPrefix(trimmed, "/v1/") ||
+		strings.HasPrefix(trimmed, "/v1beta/") ||
+		strings.HasPrefix(trimmed, "/sora/") ||
+		strings.HasPrefix(trimmed, "/antigravity/") ||
+		strings.HasPrefix(trimmed, "/setup/") ||
+		trimmed == "/health" ||
+		trimmed == "/responses" ||
+		strings.HasPrefix(trimmed, "/responses/")
+}
+
 func serveIndexHTML(c *gin.Context, fsys fs.FS) {
 	file, err := fsys.Open("index.html")
 	if err != nil {

backend/internal/web/embed_test.go

@@ -367,6 +367,7 @@ func TestFrontendServer_Middleware(t *testing.T) {
 			"/setup/init",
 			"/health",
 			"/responses",
+			"/responses/compact",
 		}
 
 		for _, path := range apiPaths {
@@ -388,6 +389,32 @@ func TestFrontendServer_Middleware(t *testing.T) {
 		}
 	})
 
+	t.Run("skips_responses_compact_post_routes", func(t *testing.T) {
+		provider := &mockSettingsProvider{
+			settings: map[string]string{"test": "value"},
+		}
+
+		server, err := NewFrontendServer(provider)
+		require.NoError(t, err)
+
+		router := gin.New()
+		router.Use(server.Middleware())
+		nextCalled := false
+		router.POST("/responses/compact", func(c *gin.Context) {
+			nextCalled = true
+			c.String(http.StatusOK, `{"ok":true}`)
+		})
+
+		w := httptest.NewRecorder()
+		req := httptest.NewRequest(http.MethodPost, "/responses/compact", strings.NewReader(`{"model":"gpt-5"}`))
+		req.Header.Set("Content-Type", "application/json")
+		router.ServeHTTP(w, req)
+
+		assert.True(t, nextCalled, "next handler should be called for compact API route")
+		assert.Equal(t, http.StatusOK, w.Code)
+		assert.JSONEq(t, `{"ok":true}`, w.Body.String())
+	})
+
 	t.Run("serves_index_for_spa_routes", func(t *testing.T) {
 		provider := &mockSettingsProvider{
 			settings: map[string]string{"test": "value"},
@@ -543,6 +570,7 @@ func TestServeEmbeddedFrontend(t *testing.T) {
 			"/setup/init",
 			"/health",
 			"/responses",
+			"/responses/compact",
 		}
 
 		for _, path := range apiPaths {

backend/migrations/068_add_announcement_notify_mode.sql

@@ -0,0 +1 @@
+ALTER TABLE announcements ADD COLUMN IF NOT EXISTS notify_mode VARCHAR(20) NOT NULL DEFAULT 'silent';

backend/migrations/069_add_group_messages_dispatch.sql

@@ -0,0 +1,2 @@
+ALTER TABLE groups ADD COLUMN allow_messages_dispatch BOOLEAN NOT NULL DEFAULT false;
+ALTER TABLE groups ADD COLUMN default_mapped_model VARCHAR(100) NOT NULL DEFAULT '';

deploy/Dockerfile

@@ -7,7 +7,7 @@
 # =============================================================================
 
 ARG NODE_IMAGE=node:24-alpine
-ARG GOLANG_IMAGE=golang:1.25.5-alpine
+ARG GOLANG_IMAGE=golang:1.26.1-alpine
 ARG ALPINE_IMAGE=alpine:3.20
 ARG GOPROXY=https://goproxy.cn,direct
 ARG GOSUMDB=sum.golang.google.cn

frontend/src/App.vue

@@ -1,9 +1,10 @@
 <script setup lang="ts">
 import { RouterView, useRouter, useRoute } from 'vue-router'
-import { onMounted, watch } from 'vue'
+import { onMounted, onBeforeUnmount, watch } from 'vue'
 import Toast from '@/components/common/Toast.vue'
 import NavigationProgress from '@/components/common/NavigationProgress.vue'
-import { useAppStore, useAuthStore, useSubscriptionStore } from '@/stores'
+import AnnouncementPopup from '@/components/common/AnnouncementPopup.vue'
+import { useAppStore, useAuthStore, useSubscriptionStore, useAnnouncementStore } from '@/stores'
 import { getSetupStatus } from '@/api/setup'
 
 const router = useRouter()
@@ -11,6 +12,7 @@ const route = useRoute()
 const appStore = useAppStore()
 const authStore = useAuthStore()
 const subscriptionStore = useSubscriptionStore()
+const announcementStore = useAnnouncementStore()
 
 /**
  * Update favicon dynamically
@@ -39,24 +41,55 @@ watch(
   { immediate: true }
 )
 
-// Watch for authentication state and manage subscription data
+// Watch for authentication state and manage subscription data + announcements
+function onVisibilityChange() {
+  if (document.visibilityState === 'visible' && authStore.isAuthenticated) {
+    announcementStore.fetchAnnouncements()
+  }
+}
+
 watch(
   () => authStore.isAuthenticated,
-  (isAuthenticated) => {
+  (isAuthenticated, oldValue) => {
     if (isAuthenticated) {
       // User logged in: preload subscriptions and start polling
       subscriptionStore.fetchActiveSubscriptions().catch((error) => {
         console.error('Failed to preload subscriptions:', error)
       })
       subscriptionStore.startPolling()
+
+      // Announcements: new login vs page refresh restore
+      if (oldValue === false) {
+        // New login: delay 3s then force fetch
+        setTimeout(() => announcementStore.fetchAnnouncements(true), 3000)
+      } else {
+        // Page refresh restore (oldValue was undefined)
+        announcementStore.fetchAnnouncements()
+      }
+
+      // Register visibility change listener
+      document.addEventListener('visibilitychange', onVisibilityChange)
     } else {
       // User logged out: clear data and stop polling
       subscriptionStore.clear()
+      announcementStore.reset()
+      document.removeEventListener('visibilitychange', onVisibilityChange)
     }
   },
   { immediate: true }
 )
 
+// Route change trigger (throttled by store)
+router.afterEach(() => {
+  if (authStore.isAuthenticated) {
+    announcementStore.fetchAnnouncements()
+  }
+})
+
+onBeforeUnmount(() => {
+  document.removeEventListener('visibilitychange', onVisibilityChange)
+})
+
 onMounted(async () => {
   // Check if setup is needed
   try {
@@ -78,4 +111,5 @@ onMounted(async () => {
   <NavigationProgress />
   <RouterView />
   <Toast />
+  <AnnouncementPopup />
 </template>

frontend/src/components/account/AccountUsageCell.vue

@@ -90,6 +90,36 @@
           color="emerald"
         />
       </div>
+      <div v-else-if="loading" class="space-y-1.5">
+        <div class="flex items-center gap-1">
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-1.5 w-8 animate-pulse rounded-full bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+        </div>
+        <div class="flex items-center gap-1">
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-1.5 w-8 animate-pulse rounded-full bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+        </div>
+      </div>
+      <div v-else-if="hasOpenAIUsageFallback" class="space-y-1">
+        <UsageProgressBar
+          v-if="usageInfo?.five_hour"
+          label="5h"
+          :utilization="usageInfo.five_hour.utilization"
+          :resets-at="usageInfo.five_hour.resets_at"
+          :window-stats="usageInfo.five_hour.window_stats"
+          color="indigo"
+        />
+        <UsageProgressBar
+          v-if="usageInfo?.seven_day"
+          label="7d"
+          :utilization="usageInfo.seven_day.utilization"
+          :resets-at="usageInfo.seven_day.resets_at"
+          :window-stats="usageInfo.seven_day.window_stats"
+          color="emerald"
+        />
+      </div>
       <div v-else class="text-xs text-gray-400">-</div>
     </template>
 
@@ -313,6 +343,9 @@ const shouldFetchUsage = computed(() => {
   if (props.account.platform === 'antigravity') {
     return props.account.type === 'oauth'
   }
+  if (props.account.platform === 'openai') {
+    return props.account.type === 'oauth'
+  }
   return false
 })
 
@@ -335,6 +368,11 @@ const hasCodexUsage = computed(() => {
   return codex5hWindow.value.usedPercent !== null || codex7dWindow.value.usedPercent !== null
 })
 
+const hasOpenAIUsageFallback = computed(() => {
+  if (props.account.platform !== 'openai' || props.account.type !== 'oauth') return false
+  return !!usageInfo.value?.five_hour || !!usageInfo.value?.seven_day
+})
+
 const codex5hUsedPercent = computed(() => codex5hWindow.value.usedPercent)
 const codex5hResetAt = computed(() => codex5hWindow.value.resetAt)
 const codex7dUsedPercent = computed(() => codex7dWindow.value.usedPercent)

frontend/src/components/account/CreateAccountModal.vue

@@ -1230,6 +1230,142 @@
       <!-- API Key 账号配额限制 -->
       <QuotaLimitCard v-if="form.type === 'apikey'" v-model="editQuotaLimit" />
 
+      <!-- OpenAI OAuth Model Mapping (OAuth 类型没有 apikey 容器，需要独立的模型映射区域) -->
+      <div
+        v-if="form.platform === 'openai' && accountCategory === 'oauth-based'"
+        class="border-t border-gray-200 pt-4 dark:border-dark-600"
+      >
+        <label class="input-label">{{ t('admin.accounts.modelRestriction') }}</label>
+
+        <div
+          v-if="isOpenAIModelRestrictionDisabled"
+          class="mb-3 rounded-lg bg-amber-50 p-3 dark:bg-amber-900/20"
+        >
+          <p class="text-xs text-amber-700 dark:text-amber-400">
+            {{ t('admin.accounts.openai.modelRestrictionDisabledByPassthrough') }}
+          </p>
+        </div>
+
+        <template v-else>
+          <!-- Mode Toggle -->
+          <div class="mb-4 flex gap-2">
+            <button
+              type="button"
+              @click="modelRestrictionMode = 'whitelist'"
+              :class="[
+                'flex-1 rounded-lg px-4 py-2 text-sm font-medium transition-all',
+                modelRestrictionMode === 'whitelist'
+                  ? 'bg-primary-100 text-primary-700 dark:bg-primary-900/30 dark:text-primary-400'
+                  : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-dark-600 dark:text-gray-400 dark:hover:bg-dark-500'
+              ]"
+            >
+              {{ t('admin.accounts.modelWhitelist') }}
+            </button>
+            <button
+              type="button"
+              @click="modelRestrictionMode = 'mapping'"
+              :class="[
+                'flex-1 rounded-lg px-4 py-2 text-sm font-medium transition-all',
+                modelRestrictionMode === 'mapping'
+                  ? 'bg-purple-100 text-purple-700 dark:bg-purple-900/30 dark:text-purple-400'
+                  : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-dark-600 dark:text-gray-400 dark:hover:bg-dark-500'
+              ]"
+            >
+              {{ t('admin.accounts.modelMapping') }}
+            </button>
+          </div>
+
+          <!-- Whitelist Mode -->
+          <div v-if="modelRestrictionMode === 'whitelist'">
+            <ModelWhitelistSelector v-model="allowedModels" :platform="form.platform" />
+            <p class="text-xs text-gray-500 dark:text-gray-400">
+              {{ t('admin.accounts.selectedModels', { count: allowedModels.length }) }}
+              <span v-if="allowedModels.length === 0">{{
+                t('admin.accounts.supportsAllModels')
+              }}</span>
+            </p>
+          </div>
+
+          <!-- Mapping Mode -->
+          <div v-else>
+            <div class="mb-3 rounded-lg bg-purple-50 p-3 dark:bg-purple-900/20">
+              <p class="text-xs text-purple-700 dark:text-purple-400">
+                {{ t('admin.accounts.mapRequestModels') }}
+              </p>
+            </div>
+
+            <div v-if="modelMappings.length > 0" class="mb-3 space-y-2">
+              <div
+                v-for="(mapping, index) in modelMappings"
+                :key="'oauth-' + getModelMappingKey(mapping)"
+                class="flex items-center gap-2"
+              >
+                <input
+                  v-model="mapping.from"
+                  type="text"
+                  class="input flex-1"
+                  :placeholder="t('admin.accounts.requestModel')"
+                />
+                <svg
+                  class="h-4 w-4 flex-shrink-0 text-gray-400"
+                  fill="none"
+                  viewBox="0 0 24 24"
+                  stroke="currentColor"
+                >
+                  <path
+                    stroke-linecap="round"
+                    stroke-linejoin="round"
+                    stroke-width="2"
+                    d="M14 5l7 7m0 0l-7 7m7-7H3"
+                  />
+                </svg>
+                <input
+                  v-model="mapping.to"
+                  type="text"
+                  class="input flex-1"
+                  :placeholder="t('admin.accounts.actualModel')"
+                />
+                <button
+                  type="button"
+                  @click="removeModelMapping(index)"
+                  class="rounded-lg p-2 text-red-500 transition-colors hover:bg-red-50 hover:text-red-600 dark:hover:bg-red-900/20"
+                >
+                  <svg class="h-4 w-4" fill="none" viewBox="0 0 24 24" stroke="currentColor">
+                    <path
+                      stroke-linecap="round"
+                      stroke-linejoin="round"
+                      stroke-width="2"
+                      d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"
+                    />
+                  </svg>
+                </button>
+              </div>
+            </div>
+
+            <button
+              type="button"
+              @click="addModelMapping"
+              class="mb-3 w-full rounded-lg border-2 border-dashed border-gray-300 px-4 py-2 text-gray-600 transition-colors hover:border-gray-400 hover:text-gray-700 dark:border-dark-500 dark:text-gray-400 dark:hover:border-dark-400 dark:hover:text-gray-300"
+            >
+              + {{ t('admin.accounts.addMapping') }}
+            </button>
+
+            <!-- Quick Add Buttons -->
+            <div class="flex flex-wrap gap-2">
+              <button
+                v-for="preset in presetMappings"
+                :key="'oauth-' + preset.label"
+                type="button"
+                @click="addPresetMapping(preset.from, preset.to)"
+                :class="['rounded-lg px-3 py-1 text-xs transition-colors', preset.color]"
+              >
+                + {{ preset.label }}
+              </button>
+            </div>
+          </div>
+        </template>
+      </div>
+
       <!-- Temp Unschedulable Rules -->
       <div class="border-t border-gray-200 pt-4 dark:border-dark-600 space-y-4">
         <div class="mb-3 flex items-center justify-between">
@@ -3603,6 +3739,14 @@ const handleOpenAIExchange = async (authCode: string) => {
     const shouldCreateOpenAI = form.platform === 'openai'
     const shouldCreateSora = form.platform === 'sora'
 
+    // Add model mapping for OpenAI OAuth accounts（透传模式下不应用）
+    if (shouldCreateOpenAI && !isOpenAIModelRestrictionDisabled.value) {
+      const modelMapping = buildModelMappingObject(modelRestrictionMode.value, allowedModels.value, modelMappings.value)
+      if (modelMapping) {
+        credentials.model_mapping = modelMapping
+      }
+    }
+
     // 应用临时不可调度配置
     if (!applyTempUnschedConfig(credentials)) {
       return
@@ -3713,6 +3857,14 @@ const handleOpenAIValidateRT = async (refreshTokenInput: string) => {
         const oauthExtra = oauthClient.buildExtraInfo(tokenInfo) as Record<string, unknown> | undefined
         const extra = buildOpenAIExtra(oauthExtra)
 
+        // Add model mapping for OpenAI OAuth accounts（透传模式下不应用）
+        if (shouldCreateOpenAI && !isOpenAIModelRestrictionDisabled.value) {
+          const modelMapping = buildModelMappingObject(modelRestrictionMode.value, allowedModels.value, modelMappings.value)
+          if (modelMapping) {
+            credentials.model_mapping = modelMapping
+          }
+        }
+
         // Generate account name with index for batch
         const accountName = refreshTokens.length > 1 ? `${form.name} #${i + 1}` : form.name
 

frontend/src/components/account/EditAccountModal.vue

@@ -351,6 +351,142 @@
 
       </div>
 
+      <!-- OpenAI OAuth Model Mapping (OAuth 类型没有 apikey 容器，需要独立的模型映射区域) -->
+      <div
+        v-if="account.platform === 'openai' && account.type === 'oauth'"
+        class="border-t border-gray-200 pt-4 dark:border-dark-600"
+      >
+        <label class="input-label">{{ t('admin.accounts.modelRestriction') }}</label>
+
+        <div
+          v-if="isOpenAIModelRestrictionDisabled"
+          class="mb-3 rounded-lg bg-amber-50 p-3 dark:bg-amber-900/20"
+        >
+          <p class="text-xs text-amber-700 dark:text-amber-400">
+            {{ t('admin.accounts.openai.modelRestrictionDisabledByPassthrough') }}
+          </p>
+        </div>
+
+        <template v-else>
+          <!-- Mode Toggle -->
+          <div class="mb-4 flex gap-2">
+            <button
+              type="button"
+              @click="modelRestrictionMode = 'whitelist'"
+              :class="[
+                'flex-1 rounded-lg px-4 py-2 text-sm font-medium transition-all',
+                modelRestrictionMode === 'whitelist'
+                  ? 'bg-primary-100 text-primary-700 dark:bg-primary-900/30 dark:text-primary-400'
+                  : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-dark-600 dark:text-gray-400 dark:hover:bg-dark-500'
+              ]"
+            >
+              {{ t('admin.accounts.modelWhitelist') }}
+            </button>
+            <button
+              type="button"
+              @click="modelRestrictionMode = 'mapping'"
+              :class="[
+                'flex-1 rounded-lg px-4 py-2 text-sm font-medium transition-all',
+                modelRestrictionMode === 'mapping'
+                  ? 'bg-purple-100 text-purple-700 dark:bg-purple-900/30 dark:text-purple-400'
+                  : 'bg-gray-100 text-gray-600 hover:bg-gray-200 dark:bg-dark-600 dark:text-gray-400 dark:hover:bg-dark-500'
+              ]"
+            >
+              {{ t('admin.accounts.modelMapping') }}
+            </button>
+          </div>
+
+          <!-- Whitelist Mode -->
+          <div v-if="modelRestrictionMode === 'whitelist'">
+            <ModelWhitelistSelector v-model="allowedModels" :platform="account?.platform || 'anthropic'" />
+            <p class="text-xs text-gray-500 dark:text-gray-400">
+              {{ t('admin.accounts.selectedModels', { count: allowedModels.length }) }}
+              <span v-if="allowedModels.length === 0">{{
+                t('admin.accounts.supportsAllModels')
+              }}</span>
+            </p>
+          </div>
+
+          <!-- Mapping Mode -->
+          <div v-else>
+            <div class="mb-3 rounded-lg bg-purple-50 p-3 dark:bg-purple-900/20">
+              <p class="text-xs text-purple-700 dark:text-purple-400">
+                {{ t('admin.accounts.mapRequestModels') }}
+              </p>
+            </div>
+
+            <div v-if="modelMappings.length > 0" class="mb-3 space-y-2">
+              <div
+                v-for="(mapping, index) in modelMappings"
+                :key="'oauth-' + getModelMappingKey(mapping)"
+                class="flex items-center gap-2"
+              >
+                <input
+                  v-model="mapping.from"
+                  type="text"
+                  class="input flex-1"
+                  :placeholder="t('admin.accounts.requestModel')"
+                />
+                <svg
+                  class="h-4 w-4 flex-shrink-0 text-gray-400"
+                  fill="none"
+                  viewBox="0 0 24 24"
+                  stroke="currentColor"
+                >
+                  <path
+                    stroke-linecap="round"
+                    stroke-linejoin="round"
+                    stroke-width="2"
+                    d="M14 5l7 7m0 0l-7 7m7-7H3"
+                  />
+                </svg>
+                <input
+                  v-model="mapping.to"
+                  type="text"
+                  class="input flex-1"
+                  :placeholder="t('admin.accounts.actualModel')"
+                />
+                <button
+                  type="button"
+                  @click="removeModelMapping(index)"
+                  class="rounded-lg p-2 text-red-500 transition-colors hover:bg-red-50 hover:text-red-600 dark:hover:bg-red-900/20"
+                >
+                  <svg class="h-4 w-4" fill="none" viewBox="0 0 24 24" stroke="currentColor">
+                    <path
+                      stroke-linecap="round"
+                      stroke-linejoin="round"
+                      stroke-width="2"
+                      d="M19 7l-.867 12.142A2 2 0 0116.138 21H7.862a2 2 0 01-1.995-1.858L5 7m5 4v6m4-6v6m1-10V4a1 1 0 00-1-1h-4a1 1 0 00-1 1v3M4 7h16"
+                    />
+                  </svg>
+                </button>
+              </div>
+            </div>
+
+            <button
+              type="button"
+              @click="addModelMapping"
+              class="mb-3 w-full rounded-lg border-2 border-dashed border-gray-300 px-4 py-2 text-gray-600 transition-colors hover:border-gray-400 hover:text-gray-700 dark:border-dark-500 dark:text-gray-400 dark:hover:border-dark-400 dark:hover:text-gray-300"
+            >
+              + {{ t('admin.accounts.addMapping') }}
+            </button>
+
+            <!-- Quick Add Buttons -->
+            <div class="flex flex-wrap gap-2">
+              <button
+                v-for="preset in presetMappings"
+                :key="'oauth-' + preset.label"
+                type="button"
+                @click="addPresetMapping(preset.from, preset.to)"
+                :class="['rounded-lg px-3 py-1 text-xs transition-colors', preset.color]"
+              >
+                + {{ preset.label }}
+              </button>
+            </div>
+          </div>
+        </template>
+      </div>
+
       <!-- Upstream fields (only for upstream type) -->
       <div v-if="account.type === 'upstream'" class="space-y-4">
         <div>
@@ -1481,15 +1617,21 @@ const form = reactive({
   load_factor: null as number | null,
   priority: 1,
   rate_multiplier: 1,
-  status: 'active' as 'active' | 'inactive',
+  status: 'active' as 'active' | 'inactive' | 'error',
   group_ids: [] as number[],
   expires_at: null as number | null
 })
 
-const statusOptions = computed(() => [
-  { value: 'active', label: t('common.active') },
-  { value: 'inactive', label: t('common.inactive') }
-])
+const statusOptions = computed(() => {
+  const options = [
+    { value: 'active', label: t('common.active') },
+    { value: 'inactive', label: t('common.inactive') }
+  ]
+  if (form.status === 'error') {
+    options.push({ value: 'error', label: t('admin.accounts.status.error') })
+  }
+  return options
+})
 
 const expiresAtInput = computed({
   get: () => formatDateTimeLocal(form.expires_at),
@@ -1515,7 +1657,7 @@ watch(
       form.load_factor = newAccount.load_factor ?? null
       form.priority = newAccount.priority
       form.rate_multiplier = newAccount.rate_multiplier ?? 1
-      form.status = (newAccount.status === 'active' || newAccount.status === 'inactive')
+      form.status = (newAccount.status === 'active' || newAccount.status === 'inactive' || newAccount.status === 'error')
         ? newAccount.status
         : 'active'
       form.group_ids = newAccount.group_ids || []
@@ -1659,9 +1801,33 @@ watch(
               ? 'https://generativelanguage.googleapis.com'
               : 'https://api.anthropic.com'
         editBaseUrl.value = platformDefaultUrl
-        modelRestrictionMode.value = 'whitelist'
-        modelMappings.value = []
-        allowedModels.value = []
+
+        // Load model mappings for OpenAI OAuth accounts
+        if (newAccount.platform === 'openai' && newAccount.credentials) {
+          const oauthCredentials = newAccount.credentials as Record<string, unknown>
+          const existingMappings = oauthCredentials.model_mapping as Record<string, string> | undefined
+          if (existingMappings && typeof existingMappings === 'object') {
+            const entries = Object.entries(existingMappings)
+            const isWhitelistMode = entries.length > 0 && entries.every(([from, to]) => from === to)
+            if (isWhitelistMode) {
+              modelRestrictionMode.value = 'whitelist'
+              allowedModels.value = entries.map(([from]) => from)
+              modelMappings.value = []
+            } else {
+              modelRestrictionMode.value = 'mapping'
+              modelMappings.value = entries.map(([from, to]) => ({ from, to }))
+              allowedModels.value = []
+            }
+          } else {
+            modelRestrictionMode.value = 'whitelist'
+            modelMappings.value = []
+            allowedModels.value = []
+          }
+        } else {
+          modelRestrictionMode.value = 'whitelist'
+          modelMappings.value = []
+          allowedModels.value = []
+        }
         customErrorCodesEnabled.value = false
         selectedErrorCodes.value = []
       }
@@ -2065,7 +2231,7 @@ const handleSubmit = async () => {
   if (!props.account) return
   const accountID = props.account.id
 
-  if (form.status !== 'active' && form.status !== 'inactive') {
+  if (form.status !== 'active' && form.status !== 'inactive' && form.status !== 'error') {
     appStore.showError(t('admin.accounts.pleaseSelectStatus'))
     return
   }
@@ -2094,6 +2260,7 @@ const handleSubmit = async () => {
 
       // Always update credentials for apikey type to handle model mapping changes
       const newCredentials: Record<string, unknown> = {
+        ...currentCredentials,
         base_url: newBaseUrl
       }
 
@@ -2114,6 +2281,8 @@ const handleSubmit = async () => {
         const modelMapping = buildModelMappingObject(modelRestrictionMode.value, allowedModels.value, modelMappings.value)
         if (modelMapping) {
           newCredentials.model_mapping = modelMapping
+        } else {
+          delete newCredentials.model_mapping
         }
       } else if (currentCredentials.model_mapping) {
         newCredentials.model_mapping = currentCredentials.model_mapping
@@ -2123,6 +2292,9 @@ const handleSubmit = async () => {
       if (customErrorCodesEnabled.value) {
         newCredentials.custom_error_codes_enabled = true
         newCredentials.custom_error_codes = [...selectedErrorCodes.value]
+      } else {
+        delete newCredentials.custom_error_codes_enabled
+        delete newCredentials.custom_error_codes
       }
 
       // Add intercept warmup requests setting
@@ -2163,6 +2335,28 @@ const handleSubmit = async () => {
       updatePayload.credentials = newCredentials
     }
 
+    // OpenAI OAuth: persist model mapping to credentials
+    if (props.account.platform === 'openai' && props.account.type === 'oauth') {
+      const currentCredentials = (updatePayload.credentials as Record<string, unknown>) ||
+        ((props.account.credentials as Record<string, unknown>) || {})
+      const newCredentials: Record<string, unknown> = { ...currentCredentials }
+      const shouldApplyModelMapping = !openaiPassthroughEnabled.value
+
+      if (shouldApplyModelMapping) {
+        const modelMapping = buildModelMappingObject(modelRestrictionMode.value, allowedModels.value, modelMappings.value)
+        if (modelMapping) {
+          newCredentials.model_mapping = modelMapping
+        } else {
+          delete newCredentials.model_mapping
+        }
+      } else if (currentCredentials.model_mapping) {
+        // 透传模式保留现有映射
+        newCredentials.model_mapping = currentCredentials.model_mapping
+      }
+
+      updatePayload.credentials = newCredentials
+    }
+
     // Antigravity: persist model mapping to credentials (applies to all antigravity types)
     // Antigravity 只支持映射模式
     if (props.account.platform === 'antigravity') {

frontend/src/components/account/__tests__/AccountUsageCell.spec.ts

@@ -67,4 +67,59 @@ describe('AccountUsageCell', () => {
 
     expect(wrapper.text()).toContain('admin.accounts.usageWindow.gemini3Image|70|2026-03-01T09:00:00Z')
   })
+
+  it('OpenAI OAuth 在无 codex 快照时会回退显示 usage 接口窗口', async () => {
+	getUsage.mockResolvedValue({
+	  five_hour: {
+	    utilization: 0,
+	    resets_at: null,
+	    remaining_seconds: 0,
+	    window_stats: {
+	      requests: 2,
+	      tokens: 27700,
+	      cost: 0.06,
+	      standard_cost: 0.06,
+	      user_cost: 0.06
+	    }
+	  },
+	  seven_day: {
+	    utilization: 0,
+	    resets_at: null,
+	    remaining_seconds: 0,
+	    window_stats: {
+	      requests: 2,
+	      tokens: 27700,
+	      cost: 0.06,
+	      standard_cost: 0.06,
+	      user_cost: 0.06
+	    }
+	  }
+	})
+
+	const wrapper = mount(AccountUsageCell, {
+	  props: {
+	    account: {
+	      id: 2002,
+	      platform: 'openai',
+	      type: 'oauth',
+	      extra: {}
+	    } as any
+	  },
+	  global: {
+	    stubs: {
+	      UsageProgressBar: {
+	        props: ['label', 'utilization', 'resetsAt', 'windowStats', 'color'],
+	        template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+	      },
+	      AccountQuotaInfo: true
+	    }
+	  }
+	})
+
+	await flushPromises()
+
+	expect(getUsage).toHaveBeenCalledWith(2002)
+	expect(wrapper.text()).toContain('5h|0|27700')
+	expect(wrapper.text()).toContain('7d|0|27700')
+  })
 })

frontend/src/components/common/AnnouncementBell.vue

@@ -314,16 +314,18 @@
 <script setup lang="ts">
 import { ref, computed, onMounted, onBeforeUnmount, watch } from 'vue'
 import { useI18n } from 'vue-i18n'
+import { storeToRefs } from 'pinia'
 import { marked } from 'marked'
 import DOMPurify from 'dompurify'
-import { announcementsAPI } from '@/api'
 import { useAppStore } from '@/stores/app'
+import { useAnnouncementStore } from '@/stores/announcements'
 import { formatRelativeTime, formatRelativeWithDateTime } from '@/utils/format'
 import type { UserAnnouncement } from '@/types'
 import Icon from '@/components/icons/Icon.vue'
 
 const { t } = useI18n()
 const appStore = useAppStore()
+const announcementStore = useAnnouncementStore()
 
 // Configure marked
 marked.setOptions({
@@ -331,17 +333,14 @@ marked.setOptions({
   gfm: true,
 })
 
-// State
-const announcements = ref<UserAnnouncement[]>([])
+// Use store state (storeToRefs for reactivity)
+const { announcements, loading } = storeToRefs(announcementStore)
+const unreadCount = computed(() => announcementStore.unreadCount)
+
+// Local modal state
 const isModalOpen = ref(false)
 const detailModalOpen = ref(false)
 const selectedAnnouncement = ref<UserAnnouncement | null>(null)
-const loading = ref(false)
-
-// Computed
-const unreadCount = computed(() =>
-  announcements.value.filter((a) => !a.read_at).length
-)
 
 // Methods
 function renderMarkdown(content: string): string {
@@ -350,24 +349,8 @@ function renderMarkdown(content: string): string {
   return DOMPurify.sanitize(html)
 }
 
-async function loadAnnouncements() {
-  try {
-    loading.value = true
-    const allAnnouncements = await announcementsAPI.list(false)
-    announcements.value = allAnnouncements.slice(0, 20)
-  } catch (err: any) {
-    console.error('Failed to load announcements:', err)
-    appStore.showError(err?.message || t('common.unknownError'))
-  } finally {
-    loading.value = false
-  }
-}
-
 function openModal() {
   isModalOpen.value = true
-  if (announcements.value.length === 0) {
-    loadAnnouncements()
-  }
 }
 
 function closeModal() {
@@ -389,14 +372,7 @@ function closeDetail() {
 
 async function markAsRead(id: number) {
   try {
-    await announcementsAPI.markRead(id)
-    const announcement = announcements.value.find((a) => a.id === id)
-    if (announcement) {
-      announcement.read_at = new Date().toISOString()
-    }
-    if (selectedAnnouncement.value?.id === id) {
-      selectedAnnouncement.value.read_at = new Date().toISOString()
-    }
+    await announcementStore.markAsRead(id)
   } catch (err: any) {
     appStore.showError(err?.message || t('common.unknownError'))
   }
@@ -410,19 +386,10 @@ async function markAsReadAndClose(id: number) {
 
 async function markAllAsRead() {
   try {
-    loading.value = true
-    const unreadAnnouncements = announcements.value.filter((a) => !a.read_at)
-    await Promise.all(unreadAnnouncements.map((a) => announcementsAPI.markRead(a.id)))
-    announcements.value.forEach((a) => {
-      if (!a.read_at) {
-        a.read_at = new Date().toISOString()
-      }
-    })
+    await announcementStore.markAllAsRead()
     appStore.showSuccess(t('announcements.allMarkedAsRead'))
   } catch (err: any) {
     appStore.showError(err?.message || t('common.unknownError'))
-  } finally {
-    loading.value = false
   }
 }
 
@@ -438,22 +405,19 @@ function handleEscape(e: KeyboardEvent) {
 
 onMounted(() => {
   document.addEventListener('keydown', handleEscape)
-  loadAnnouncements()
 })
 
 onBeforeUnmount(() => {
   document.removeEventListener('keydown', handleEscape)
-  // Restore body overflow in case component is unmounted while modals are open
   document.body.style.overflow = ''
 })
 
-watch([isModalOpen, detailModalOpen], ([modal, detail]) => {
-  if (modal || detail) {
-    document.body.style.overflow = 'hidden'
-  } else {
-    document.body.style.overflow = ''
+watch(
+  [isModalOpen, detailModalOpen, () => announcementStore.currentPopup],
+  ([modal, detail, popup]) => {
+    document.body.style.overflow = (modal || detail || popup) ? 'hidden' : ''
   }
-})
+)
 </script>
 
 <style scoped>

frontend/src/components/common/AnnouncementPopup.vue

@@ -0,0 +1,165 @@
+<template>
+  <Teleport to="body">
+    <Transition name="popup-fade">
+      <div
+        v-if="announcementStore.currentPopup"
+        class="fixed inset-0 z-[120] flex items-start justify-center overflow-y-auto bg-gradient-to-br from-black/70 via-black/60 to-black/70 p-4 pt-[8vh] backdrop-blur-md"
+      >
+        <div
+          class="w-full max-w-[680px] overflow-hidden rounded-3xl bg-white shadow-2xl ring-1 ring-black/5 dark:bg-dark-800 dark:ring-white/10"
+          @click.stop
+        >
+          <!-- Header with warm gradient -->
+          <div class="relative overflow-hidden border-b border-amber-100/80 bg-gradient-to-br from-amber-50/80 via-orange-50/50 to-yellow-50/30 px-8 py-6 dark:border-dark-700/50 dark:from-amber-900/20 dark:via-orange-900/10 dark:to-yellow-900/5">
+            <!-- Decorative background -->
+            <div class="absolute right-0 top-0 h-full w-64 bg-gradient-to-l from-orange-100/30 to-transparent dark:from-orange-900/20"></div>
+            <div class="absolute -right-8 -top-8 h-32 w-32 rounded-full bg-gradient-to-br from-amber-400/20 to-orange-500/20 blur-3xl"></div>
+            <div class="absolute -left-4 -bottom-4 h-24 w-24 rounded-full bg-gradient-to-tr from-yellow-400/20 to-amber-500/20 blur-2xl"></div>
+
+            <div class="relative z-10">
+              <!-- Icon and badge -->
+              <div class="mb-3 flex items-center gap-2">
+                <div class="flex h-10 w-10 items-center justify-center rounded-xl bg-gradient-to-br from-amber-500 to-orange-600 text-white shadow-lg shadow-amber-500/30">
+                  <svg class="h-5 w-5" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                    <path stroke-linecap="round" stroke-linejoin="round" d="M15 17h5l-1.405-1.405A2.032 2.032 0 0118 14.158V11a6.002 6.002 0 00-4-5.659V5a2 2 0 10-4 0v.341C7.67 6.165 6 8.388 6 11v3.159c0 .538-.214 1.055-.595 1.436L4 17h5m6 0v1a3 3 0 11-6 0v-1m6 0H9" />
+                  </svg>
+                </div>
+                <span class="inline-flex items-center gap-1.5 rounded-lg bg-gradient-to-r from-amber-500 to-orange-600 px-2.5 py-1 text-xs font-medium text-white shadow-lg shadow-amber-500/30">
+                  <span class="relative flex h-2 w-2">
+                    <span class="absolute inline-flex h-full w-full animate-ping rounded-full bg-white opacity-75"></span>
+                    <span class="relative inline-flex h-2 w-2 rounded-full bg-white"></span>
+                  </span>
+                  {{ t('announcements.unread') }}
+                </span>
+              </div>
+
+              <!-- Title -->
+              <h2 class="mb-2 text-2xl font-bold leading-tight text-gray-900 dark:text-white">
+                {{ announcementStore.currentPopup.title }}
+              </h2>
+
+              <!-- Time -->
+              <div class="flex items-center gap-1.5 text-sm text-gray-600 dark:text-gray-400">
+                <svg class="h-4 w-4" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                  <path stroke-linecap="round" stroke-linejoin="round" d="M12 8v4l3 3m6-3a9 9 0 11-18 0 9 9 0 0118 0z" />
+                </svg>
+                <time>{{ formatRelativeWithDateTime(announcementStore.currentPopup.created_at) }}</time>
+              </div>
+            </div>
+          </div>
+
+          <!-- Body -->
+          <div class="max-h-[50vh] overflow-y-auto bg-white px-8 py-8 dark:bg-dark-800">
+            <div class="relative">
+              <div class="absolute left-0 top-0 bottom-0 w-1 rounded-full bg-gradient-to-b from-amber-500 via-orange-500 to-yellow-500"></div>
+              <div class="pl-6">
+                <div
+                  class="markdown-body prose prose-sm max-w-none dark:prose-invert"
+                  v-html="renderedContent"
+                ></div>
+              </div>
+            </div>
+          </div>
+
+          <!-- Footer -->
+          <div class="border-t border-gray-100 bg-gray-50/50 px-8 py-5 dark:border-dark-700 dark:bg-dark-900/30">
+            <div class="flex items-center justify-end">
+              <button
+                @click="handleDismiss"
+                class="rounded-xl bg-gradient-to-r from-amber-500 to-orange-600 px-6 py-2.5 text-sm font-medium text-white shadow-lg shadow-amber-500/30 transition-all hover:shadow-xl hover:scale-105"
+              >
+                <span class="flex items-center gap-2">
+                  <svg class="h-4 w-4" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                    <path stroke-linecap="round" stroke-linejoin="round" d="M5 13l4 4L19 7" />
+                  </svg>
+                  {{ t('announcements.markRead') }}
+                </span>
+              </button>
+            </div>
+          </div>
+        </div>
+      </div>
+    </Transition>
+  </Teleport>
+</template>
+
+<script setup lang="ts">
+import { computed, watch } from 'vue'
+import { useI18n } from 'vue-i18n'
+import { marked } from 'marked'
+import DOMPurify from 'dompurify'
+import { useAnnouncementStore } from '@/stores/announcements'
+import { formatRelativeWithDateTime } from '@/utils/format'
+
+const { t } = useI18n()
+const announcementStore = useAnnouncementStore()
+
+marked.setOptions({
+  breaks: true,
+  gfm: true,
+})
+
+const renderedContent = computed(() => {
+  const content = announcementStore.currentPopup?.content
+  if (!content) return ''
+  const html = marked.parse(content) as string
+  return DOMPurify.sanitize(html)
+})
+
+function handleDismiss() {
+  announcementStore.dismissPopup()
+}
+
+// Manage body overflow — only set, never unset (bell component handles restore)
+watch(
+  () => announcementStore.currentPopup,
+  (popup) => {
+    if (popup) {
+      document.body.style.overflow = 'hidden'
+    }
+  }
+)
+</script>
+
+<style scoped>
+.popup-fade-enter-active {
+  transition: all 0.3s cubic-bezier(0.16, 1, 0.3, 1);
+}
+
+.popup-fade-leave-active {
+  transition: all 0.2s cubic-bezier(0.4, 0, 1, 1);
+}
+
+.popup-fade-enter-from,
+.popup-fade-leave-to {
+  opacity: 0;
+}
+
+.popup-fade-enter-from > div {
+  transform: scale(0.94) translateY(-12px);
+  opacity: 0;
+}
+
+.popup-fade-leave-to > div {
+  transform: scale(0.96) translateY(-8px);
+  opacity: 0;
+}
+
+/* Scrollbar Styling */
+.overflow-y-auto::-webkit-scrollbar {
+  width: 8px;
+}
+
+.overflow-y-auto::-webkit-scrollbar-track {
+  background: transparent;
+}
+
+.overflow-y-auto::-webkit-scrollbar-thumb {
+  background: linear-gradient(to bottom, #cbd5e1, #94a3b8);
+  border-radius: 4px;
+}
+
+.dark .overflow-y-auto::-webkit-scrollbar-thumb {
+  background: linear-gradient(to bottom, #4b5563, #374151);
+}
+</style>

frontend/src/components/keys/UseKeyModal.vue

@@ -146,6 +146,7 @@ interface Props {
   apiKey: string
   baseUrl: string
   platform: GroupPlatform | null
+  allowMessagesDispatch?: boolean
 }
 
 interface Emits {
@@ -265,12 +266,17 @@ const SparkleIcon = {
 const clientTabs = computed((): TabConfig[] => {
   if (!props.platform) return []
   switch (props.platform) {
-    case 'openai':
-      return [
+    case 'openai': {
+      const tabs: TabConfig[] = [
         { id: 'codex', label: t('keys.useKeyModal.cliTabs.codexCli'), icon: TerminalIcon },
         { id: 'codex-ws', label: t('keys.useKeyModal.cliTabs.codexCliWs'), icon: TerminalIcon },
-        { id: 'opencode', label: t('keys.useKeyModal.cliTabs.opencode'), icon: TerminalIcon }
       ]
+      if (props.allowMessagesDispatch) {
+        tabs.push({ id: 'claude', label: t('keys.useKeyModal.cliTabs.claudeCode'), icon: TerminalIcon })
+      }
+      tabs.push({ id: 'opencode', label: t('keys.useKeyModal.cliTabs.opencode'), icon: TerminalIcon })
+      return tabs
+    }
     case 'gemini':
       return [
         { id: 'gemini', label: t('keys.useKeyModal.cliTabs.geminiCli'), icon: SparkleIcon },
@@ -316,6 +322,9 @@ const currentTabs = computed(() => {
 const platformDescription = computed(() => {
   switch (props.platform) {
     case 'openai':
+      if (activeClientTab.value === 'claude') {
+        return t('keys.useKeyModal.description')
+      }
       return t('keys.useKeyModal.openai.description')
     case 'gemini':
       return t('keys.useKeyModal.gemini.description')
@@ -329,6 +338,9 @@ const platformDescription = computed(() => {
 const platformNote = computed(() => {
   switch (props.platform) {
     case 'openai':
+      if (activeClientTab.value === 'claude') {
+        return t('keys.useKeyModal.note')
+      }
       return activeTab.value === 'windows'
         ? t('keys.useKeyModal.openai.noteWindows')
         : t('keys.useKeyModal.openai.note')
@@ -402,6 +414,9 @@ const currentFiles = computed((): FileConfig[] => {
 
   switch (props.platform) {
     case 'openai':
+      if (activeClientTab.value === 'claude') {
+        return generateAnthropicFiles(baseUrl, apiKey)
+      }
       if (activeClientTab.value === 'codex-ws') {
         return generateOpenAIWsFiles(baseUrl, apiKey)
       }

frontend/src/composables/useModelWhitelist.ts

@@ -280,7 +280,10 @@ const openaiPresetMappings = [
   { label: 'GPT-5.1', from: 'gpt-5.1', to: 'gpt-5.1', color: 'bg-orange-100 text-orange-700 hover:bg-orange-200 dark:bg-orange-900/30 dark:text-orange-400' },
   { label: 'GPT-5.2', from: 'gpt-5.2', to: 'gpt-5.2', color: 'bg-red-100 text-red-700 hover:bg-red-200 dark:bg-red-900/30 dark:text-red-400' },
   { label: 'GPT-5.4', from: 'gpt-5.4', to: 'gpt-5.4', color: 'bg-rose-100 text-rose-700 hover:bg-rose-200 dark:bg-rose-900/30 dark:text-rose-400' },
-  { label: 'GPT-5.1 Codex', from: 'gpt-5.1-codex', to: 'gpt-5.1-codex', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' }
+  { label: 'GPT-5.1 Codex', from: 'gpt-5.1-codex', to: 'gpt-5.1-codex', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' },
+  { label: 'Haiku→5.4', from: 'claude-haiku-4-5-20251001', to: 'gpt-5.4', color: 'bg-emerald-100 text-emerald-700 hover:bg-emerald-200 dark:bg-emerald-900/30 dark:text-emerald-400' },
+  { label: 'Opus→5.4', from: 'claude-opus-4-6', to: 'gpt-5.4', color: 'bg-purple-100 text-purple-700 hover:bg-purple-200 dark:bg-purple-900/30 dark:text-purple-400' },
+  { label: 'Sonnet→5.4', from: 'claude-sonnet-4-6', to: 'gpt-5.4', color: 'bg-blue-100 text-blue-700 hover:bg-blue-200 dark:bg-blue-900/30 dark:text-blue-400' }
 ]
 
 const soraPresetMappings: { label: string; from: string; to: string; color: string }[] = []

frontend/src/i18n/locales/en.ts

@@ -1443,6 +1443,14 @@ export default {
         fallbackHint: 'Non-Claude Code requests will use this group. Leave empty to reject directly.',
         noFallback: 'No Fallback (Reject)'
       },
+      openaiMessages: {
+        title: 'OpenAI Messages Dispatch',
+        allowDispatch: 'Allow /v1/messages dispatch',
+        allowDispatchHint: 'When enabled, API keys in this OpenAI group can dispatch requests through /v1/messages endpoint',
+        defaultModel: 'Default mapped model',
+        defaultModelPlaceholder: 'e.g., gpt-4.1',
+        defaultModelHint: 'When account has no model mapping configured, all request models will be mapped to this model'
+      },
       invalidRequestFallback: {
         title: 'Invalid Request Fallback Group',
         hint: 'Triggered only when upstream explicitly returns prompt too long. Leave empty to disable fallback.',
@@ -2704,6 +2712,7 @@ export default {
       columns: {
         title: 'Title',
         status: 'Status',
+        notifyMode: 'Notify Mode',
         targeting: 'Targeting',
         timeRange: 'Schedule',
         createdAt: 'Created At',
@@ -2714,10 +2723,16 @@ export default {
         active: 'Active',
         archived: 'Archived'
       },
+      notifyModeLabels: {
+        silent: 'Silent',
+        popup: 'Popup'
+      },
       form: {
         title: 'Title',
         content: 'Content (Markdown supported)',
         status: 'Status',
+        notifyMode: 'Notify Mode',
+        notifyModeHint: 'Popup mode will show a popup notification to users',
         startsAt: 'Starts At',
         endsAt: 'Ends At',
         startsAtHint: 'Leave empty to start immediately',

frontend/src/i18n/locales/zh.ts

@@ -1530,6 +1530,14 @@ export default {
         fallbackHint: '非 Claude Code 请求将使用此分组，留空则直接拒绝',
         noFallback: '不降级（直接拒绝）'
       },
+      openaiMessages: {
+        title: 'OpenAI Messages 调度配置',
+        allowDispatch: '允许 /v1/messages 调度',
+        allowDispatchHint: '启用后，此 OpenAI 分组的 API Key 可以通过 /v1/messages 端点调度请求',
+        defaultModel: '默认映射模型',
+        defaultModelPlaceholder: '例如: gpt-4.1',
+        defaultModelHint: '当账号未配置模型映射时，所有请求模型将映射到此模型'
+      },
       invalidRequestFallback: {
         title: '无效请求兜底分组',
         hint: '仅当上游明确返回 prompt too long 时才会触发，留空表示不兜底',
@@ -2640,6 +2648,12 @@ export default {
       allProtocols: '全部协议',
       allStatus: '全部状态',
       searchProxies: '搜索代理...',
+      protocols: {
+        http: 'HTTP',
+        https: 'HTTPS',
+        socks5: 'SOCKS5',
+        socks5h: 'SOCKS5H (远程 DNS)',
+      },
       name: '名称',
       protocol: '协议',
       host: '主机',
@@ -2866,6 +2880,7 @@ export default {
       columns: {
         title: '标题',
         status: '状态',
+        notifyMode: '通知方式',
         targeting: '展示条件',
         timeRange: '有效期',
         createdAt: '创建时间',
@@ -2876,10 +2891,16 @@ export default {
         active: '展示中',
         archived: '已归档'
       },
+      notifyModeLabels: {
+        silent: '静默',
+        popup: '弹窗'
+      },
       form: {
         title: '标题',
         content: '内容（支持 Markdown）',
         status: '状态',
+        notifyMode: '通知方式',
+        notifyModeHint: '弹窗模式会自动弹出通知给用户',
         startsAt: '开始时间',
         endsAt: '结束时间',
         startsAtHint: '留空表示立即生效',

frontend/src/stores/announcements.ts

@@ -0,0 +1,143 @@
+import { defineStore } from 'pinia'
+import { ref, computed } from 'vue'
+import { announcementsAPI } from '@/api'
+import type { UserAnnouncement } from '@/types'
+
+const THROTTLE_MS = 20 * 60 * 1000 // 20 minutes
+
+export const useAnnouncementStore = defineStore('announcements', () => {
+  // State
+  const announcements = ref<UserAnnouncement[]>([])
+  const loading = ref(false)
+  const lastFetchTime = ref(0)
+  const popupQueue = ref<UserAnnouncement[]>([])
+  const currentPopup = ref<UserAnnouncement | null>(null)
+
+  // Session-scoped dedup set — not reactive, used as plain lookup only
+  let shownPopupIds = new Set<number>()
+
+  // Getters
+  const unreadCount = computed(() =>
+    announcements.value.filter((a) => !a.read_at).length
+  )
+
+  // Actions
+  async function fetchAnnouncements(force = false) {
+    const now = Date.now()
+    if (!force && lastFetchTime.value > 0 && now - lastFetchTime.value < THROTTLE_MS) {
+      return
+    }
+
+    // Set immediately to prevent concurrent duplicate requests
+    lastFetchTime.value = now
+
+    try {
+      loading.value = true
+      const all = await announcementsAPI.list(false)
+      announcements.value = all.slice(0, 20)
+      enqueueNewPopups()
+    } catch (err: any) {
+      // Revert throttle timestamp on failure so retry is allowed
+      lastFetchTime.value = 0
+      console.error('Failed to fetch announcements:', err)
+    } finally {
+      loading.value = false
+    }
+  }
+
+  function enqueueNewPopups() {
+    const newPopups = announcements.value.filter(
+      (a) => a.notify_mode === 'popup' && !a.read_at && !shownPopupIds.has(a.id)
+    )
+    if (newPopups.length === 0) return
+
+    for (const p of newPopups) {
+      if (!popupQueue.value.some((q) => q.id === p.id)) {
+        popupQueue.value.push(p)
+      }
+    }
+
+    if (!currentPopup.value) {
+      showNextPopup()
+    }
+  }
+
+  function showNextPopup() {
+    if (popupQueue.value.length === 0) {
+      currentPopup.value = null
+      return
+    }
+    currentPopup.value = popupQueue.value.shift()!
+    shownPopupIds.add(currentPopup.value.id)
+  }
+
+  async function dismissPopup() {
+    if (!currentPopup.value) return
+    const id = currentPopup.value.id
+    currentPopup.value = null
+
+    // Mark as read (fire-and-forget, UI already updated)
+    markAsRead(id)
+
+    // Show next popup after a short delay
+    if (popupQueue.value.length > 0) {
+      setTimeout(() => showNextPopup(), 300)
+    }
+  }
+
+  async function markAsRead(id: number) {
+    try {
+      await announcementsAPI.markRead(id)
+      const ann = announcements.value.find((a) => a.id === id)
+      if (ann) {
+        ann.read_at = new Date().toISOString()
+      }
+    } catch (err: any) {
+      console.error('Failed to mark announcement as read:', err)
+    }
+  }
+
+  async function markAllAsRead() {
+    const unread = announcements.value.filter((a) => !a.read_at)
+    if (unread.length === 0) return
+
+    try {
+      loading.value = true
+      await Promise.all(unread.map((a) => announcementsAPI.markRead(a.id)))
+      announcements.value.forEach((a) => {
+        if (!a.read_at) {
+          a.read_at = new Date().toISOString()
+        }
+      })
+    } catch (err: any) {
+      console.error('Failed to mark all as read:', err)
+      throw err
+    } finally {
+      loading.value = false
+    }
+  }
+
+  function reset() {
+    announcements.value = []
+    lastFetchTime.value = 0
+    shownPopupIds = new Set()
+    popupQueue.value = []
+    currentPopup.value = null
+    loading.value = false
+  }
+
+  return {
+    // State
+    announcements,
+    loading,
+    currentPopup,
+    // Getters
+    unreadCount,
+    // Actions
+    fetchAnnouncements,
+    dismissPopup,
+    markAsRead,
+    markAllAsRead,
+    reset,
+  }
+})

frontend/src/stores/index.ts

@@ -8,6 +8,7 @@ export { useAppStore } from './app'
 export { useAdminSettingsStore } from './adminSettings'
 export { useSubscriptionStore } from './subscriptions'
 export { useOnboardingStore } from './onboarding'
+export { useAnnouncementStore } from './announcements'
 
 // Re-export types for convenience
 export type { User, LoginRequest, RegisterRequest, AuthResponse } from '@/types'

frontend/src/types/index.ts

@@ -155,6 +155,7 @@ export interface UpdateSubscriptionRequest {
 // ==================== Announcement Types ====================
 
 export type AnnouncementStatus = 'draft' | 'active' | 'archived'
+export type AnnouncementNotifyMode = 'silent' | 'popup'
 
 export type AnnouncementConditionType = 'subscription' | 'balance'
 
@@ -180,6 +181,7 @@ export interface Announcement {
   title: string
   content: string
   status: AnnouncementStatus
+  notify_mode: AnnouncementNotifyMode
   targeting: AnnouncementTargeting
   starts_at?: string
   ends_at?: string
@@ -193,6 +195,7 @@ export interface UserAnnouncement {
   id: number
   title: string
   content: string
+  notify_mode: AnnouncementNotifyMode
   starts_at?: string
   ends_at?: string
   read_at?: string
@@ -204,6 +207,7 @@ export interface CreateAnnouncementRequest {
   title: string
   content: string
   status?: AnnouncementStatus
+  notify_mode?: AnnouncementNotifyMode
   targeting: AnnouncementTargeting
   starts_at?: number
   ends_at?: number
@@ -213,6 +217,7 @@ export interface UpdateAnnouncementRequest {
   title?: string
   content?: string
   status?: AnnouncementStatus
+  notify_mode?: AnnouncementNotifyMode
   targeting?: AnnouncementTargeting
   starts_at?: number
   ends_at?: number
@@ -384,6 +389,8 @@ export interface Group {
   claude_code_only: boolean
   fallback_group_id: number | null
   fallback_group_id_on_invalid_request: number | null
+  // OpenAI Messages 调度开关（用户侧需要此字段判断是否展示 Claude Code 教程）
+  allow_messages_dispatch?: boolean
   created_at: string
   updated_at: string
 }
@@ -402,6 +409,9 @@ export interface AdminGroup extends Group {
   // 分组下账号数量（仅管理员可见）
   account_count?: number
 
+  // OpenAI Messages 调度配置（仅 openai 平台使用）
+  default_mapped_model?: string
+
   // 分组排序
   sort_order: number
 }
@@ -809,7 +819,7 @@ export interface UpdateAccountRequest {
   priority?: number
   rate_multiplier?: number // Account billing multiplier (>=0, 0 means free)
   schedulable?: boolean
-  status?: 'active' | 'inactive'
+  status?: 'active' | 'inactive' | 'error'
   group_ids?: number[]
   expires_at?: number | null
   auto_pause_on_expired?: boolean

frontend/src/views/admin/AnnouncementsView.vue

@@ -68,6 +68,19 @@
             </span>
           </template>
 
+          <template #cell-notifyMode="{ row }">
+            <span
+              :class="[
+                'badge',
+                row.notify_mode === 'popup'
+                  ? 'badge-warning'
+                  : 'badge-gray'
+              ]"
+            >
+              {{ row.notify_mode === 'popup' ? t('admin.announcements.notifyModeLabels.popup') : t('admin.announcements.notifyModeLabels.silent') }}
+            </span>
+          </template>
+
           <template #cell-targeting="{ row }">
             <span class="text-sm text-gray-600 dark:text-gray-300">
               {{ targetingSummary(row.targeting) }}
@@ -163,7 +176,11 @@
             <label class="input-label">{{ t('admin.announcements.form.status') }}</label>
             <Select v-model="form.status" :options="statusOptions" />
           </div>
-          <div></div>
+          <div>
+            <label class="input-label">{{ t('admin.announcements.form.notifyMode') }}</label>
+            <Select v-model="form.notify_mode" :options="notifyModeOptions" />
+            <p class="input-hint">{{ t('admin.announcements.form.notifyModeHint') }}</p>
+          </div>
         </div>
 
         <div class="grid grid-cols-1 gap-4 md:grid-cols-2">
@@ -271,9 +288,15 @@ const statusOptions = computed(() => [
   { value: 'archived', label: t('admin.announcements.statusLabels.archived') }
 ])
 
+const notifyModeOptions = computed(() => [
+  { value: 'silent', label: t('admin.announcements.notifyModeLabels.silent') },
+  { value: 'popup', label: t('admin.announcements.notifyModeLabels.popup') }
+])
+
 const columns = computed<Column[]>(() => [
   { key: 'title', label: t('admin.announcements.columns.title') },
   { key: 'status', label: t('admin.announcements.columns.status') },
+  { key: 'notifyMode', label: t('admin.announcements.columns.notifyMode') },
   { key: 'targeting', label: t('admin.announcements.columns.targeting') },
   { key: 'timeRange', label: t('admin.announcements.columns.timeRange') },
   { key: 'createdAt', label: t('admin.announcements.columns.createdAt') },
@@ -357,6 +380,7 @@ const form = reactive({
   title: '',
   content: '',
   status: 'draft',
+  notify_mode: 'silent',
   starts_at_str: '',
   ends_at_str: '',
   targeting: { any_of: [] } as AnnouncementTargeting
@@ -378,6 +402,7 @@ function resetForm() {
   form.title = ''
   form.content = ''
   form.status = 'draft'
+  form.notify_mode = 'silent'
   form.starts_at_str = ''
   form.ends_at_str = ''
   form.targeting = { any_of: [] }
@@ -387,6 +412,7 @@ function fillFormFromAnnouncement(a: Announcement) {
   form.title = a.title
   form.content = a.content
   form.status = a.status
+  form.notify_mode = a.notify_mode || 'silent'
 
   // Backend returns RFC3339 strings
   form.starts_at_str = a.starts_at ? formatDateTimeLocalInput(Math.floor(new Date(a.starts_at).getTime() / 1000)) : ''
@@ -420,6 +446,7 @@ function buildCreatePayload() {
     title: form.title,
     content: form.content,
     status: form.status as any,
+    notify_mode: form.notify_mode as any,
     targeting: form.targeting,
     starts_at: startsAt ?? undefined,
     ends_at: endsAt ?? undefined
@@ -432,6 +459,7 @@ function buildUpdatePayload(original: Announcement) {
   if (form.title !== original.title) payload.title = form.title
   if (form.content !== original.content) payload.content = form.content
   if (form.status !== original.status) payload.status = form.status
+  if (form.notify_mode !== (original.notify_mode || 'silent')) payload.notify_mode = form.notify_mode
 
   // starts_at / ends_at: distinguish unchanged vs clear(0) vs set
   const originalStarts = original.starts_at ? Math.floor(new Date(original.starts_at).getTime() / 1000) : null

frontend/src/views/admin/GroupsView.vue

@@ -708,6 +708,44 @@
           </div>
         </div>
 
+        <!-- OpenAI Messages 调度配置（仅 openai 平台） -->
+        <div v-if="createForm.platform === 'openai'" class="border-t border-gray-200 dark:border-dark-400 pt-4 mt-4">
+          <h4 class="text-sm font-medium text-gray-700 dark:text-gray-300 mb-3">{{ t('admin.groups.openaiMessages.title') }}</h4>
+
+          <!-- 允许 Messages 调度开关 -->
+          <div class="flex items-center justify-between">
+            <label class="text-sm text-gray-600 dark:text-gray-400">{{ t('admin.groups.openaiMessages.allowDispatch') }}</label>
+            <button
+              type="button"
+              @click="createForm.allow_messages_dispatch = !createForm.allow_messages_dispatch"
+              class="relative inline-flex h-6 w-12 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none"
+              :class="
+                createForm.allow_messages_dispatch ? 'bg-primary-500' : 'bg-gray-300 dark:bg-dark-600'
+              "
+            >
+              <span
+                class="pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out"
+                :class="
+                  createForm.allow_messages_dispatch ? 'translate-x-6' : 'translate-x-1'
+                "
+              />
+            </button>
+          </div>
+          <p class="text-xs text-gray-500 dark:text-gray-400 mt-1">{{ t('admin.groups.openaiMessages.allowDispatchHint') }}</p>
+
+          <!-- 默认映射模型（仅当开关打开时显示） -->
+          <div v-if="createForm.allow_messages_dispatch" class="mt-3">
+            <label class="input-label">{{ t('admin.groups.openaiMessages.defaultModel') }}</label>
+            <input
+              v-model="createForm.default_mapped_model"
+              type="text"
+              :placeholder="t('admin.groups.openaiMessages.defaultModelPlaceholder')"
+              class="input"
+            />
+            <p class="input-hint">{{ t('admin.groups.openaiMessages.defaultModelHint') }}</p>
+          </div>
+        </div>
+
         <!-- 无效请求兜底（仅 anthropic/antigravity 平台，且非订阅分组） -->
         <div
           v-if="['anthropic', 'antigravity'].includes(createForm.platform) && createForm.subscription_type !== 'subscription'"
@@ -1405,6 +1443,44 @@
           </div>
         </div>
 
+        <!-- OpenAI Messages 调度配置（仅 openai 平台） -->
+        <div v-if="editForm.platform === 'openai'" class="border-t border-gray-200 dark:border-dark-400 pt-4 mt-4">
+          <h4 class="text-sm font-medium text-gray-700 dark:text-gray-300 mb-3">{{ t('admin.groups.openaiMessages.title') }}</h4>
+
+          <!-- 允许 Messages 调度开关 -->
+          <div class="flex items-center justify-between">
+            <label class="text-sm text-gray-600 dark:text-gray-400">{{ t('admin.groups.openaiMessages.allowDispatch') }}</label>
+            <button
+              type="button"
+              @click="editForm.allow_messages_dispatch = !editForm.allow_messages_dispatch"
+              class="relative inline-flex h-6 w-12 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none"
+              :class="
+                editForm.allow_messages_dispatch ? 'bg-primary-500' : 'bg-gray-300 dark:bg-dark-600'
+              "
+            >
+              <span
+                class="pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out"
+                :class="
+                  editForm.allow_messages_dispatch ? 'translate-x-6' : 'translate-x-1'
+                "
+              />
+            </button>
+          </div>
+          <p class="text-xs text-gray-500 dark:text-gray-400 mt-1">{{ t('admin.groups.openaiMessages.allowDispatchHint') }}</p>
+
+          <!-- 默认映射模型（仅当开关打开时显示） -->
+          <div v-if="editForm.allow_messages_dispatch" class="mt-3">
+            <label class="input-label">{{ t('admin.groups.openaiMessages.defaultModel') }}</label>
+            <input
+              v-model="editForm.default_mapped_model"
+              type="text"
+              :placeholder="t('admin.groups.openaiMessages.defaultModelPlaceholder')"
+              class="input"
+            />
+            <p class="input-hint">{{ t('admin.groups.openaiMessages.defaultModelHint') }}</p>
+          </div>
+        </div>
+
         <!-- 无效请求兜底（仅 anthropic/antigravity 平台，且非订阅分组） -->
         <div
           v-if="['anthropic', 'antigravity'].includes(editForm.platform) && editForm.subscription_type !== 'subscription'"
@@ -1920,6 +1996,9 @@ const createForm = reactive({
   claude_code_only: false,
   fallback_group_id: null as number | null,
   fallback_group_id_on_invalid_request: null as number | null,
+  // OpenAI Messages 调度配置（仅 openai 平台使用）
+  allow_messages_dispatch: false,
+  default_mapped_model: 'gpt-5.4',
   // 模型路由开关
   model_routing_enabled: false,
   // 支持的模型系列（仅 antigravity 平台）
@@ -2161,6 +2240,9 @@ const editForm = reactive({
   claude_code_only: false,
   fallback_group_id: null as number | null,
   fallback_group_id_on_invalid_request: null as number | null,
+  // OpenAI Messages 调度配置（仅 openai 平台使用）
+  allow_messages_dispatch: false,
+  default_mapped_model: '',
   // 模型路由开关
   model_routing_enabled: false,
   // 支持的模型系列（仅 antigravity 平台）
@@ -2260,6 +2342,8 @@ const closeCreateModal = () => {
   createForm.claude_code_only = false
   createForm.fallback_group_id = null
   createForm.fallback_group_id_on_invalid_request = null
+  createForm.allow_messages_dispatch = false
+  createForm.default_mapped_model = 'gpt-5.4'
   createForm.supported_model_scopes = ['claude', 'gemini_text', 'gemini_image']
   createForm.mcp_xml_inject = true
   createForm.copy_accounts_from_group_ids = []
@@ -2320,6 +2404,8 @@ const handleEdit = async (group: AdminGroup) => {
   editForm.claude_code_only = group.claude_code_only || false
   editForm.fallback_group_id = group.fallback_group_id
   editForm.fallback_group_id_on_invalid_request = group.fallback_group_id_on_invalid_request
+  editForm.allow_messages_dispatch = group.allow_messages_dispatch || false
+  editForm.default_mapped_model = group.default_mapped_model || ''
   editForm.model_routing_enabled = group.model_routing_enabled || false
   editForm.supported_model_scopes = group.supported_model_scopes || ['claude', 'gemini_text', 'gemini_image']
   editForm.mcp_xml_inject = group.mcp_xml_inject ?? true
@@ -2410,6 +2496,10 @@ watch(
     if (!['anthropic', 'antigravity'].includes(newVal)) {
       createForm.fallback_group_id_on_invalid_request = null
     }
+    if (newVal !== 'openai') {
+      createForm.allow_messages_dispatch = false
+      createForm.default_mapped_model = ''
+    }
   }
 )
 

frontend/src/views/user/KeysView.vue

@@ -899,6 +899,7 @@
       :api-key="selectedKey?.key || ''"
       :base-url="publicSettings?.api_base_url || ''"
       :platform="selectedKey?.group?.platform || null"
+      :allow-messages-dispatch="selectedKey?.group?.allow_messages_dispatch || false"
       @close="closeUseKeyModal"
     />
 
@@ -1638,17 +1639,21 @@ const executeCcsImport = (row: ApiKey, clientType: 'claude' | 'gemini') => {
       headers: { "Authorization": "Bearer {{apiKey}}" }
     },
     extractor: function(response) {
+      const remaining = response?.remaining ?? response?.quota?.remaining ?? response?.balance;
+      const unit = response?.unit ?? response?.quota?.unit ?? "USD";
       return {
-        isValid: response.is_active || true,
-        remaining: response.balance,
-        unit: "USD"
+        isValid: response?.is_active ?? response?.isValid ?? true,
+        remaining,
+        unit
       };
     }
   })`
+  const providerName = (publicSettings.value?.site_name || 'sub2api').trim() || 'sub2api'
+
   const params = new URLSearchParams({
     resource: 'provider',
     app: app,
-    name: 'sub2api',
+    name: providerName,
     homepage: baseUrl,
     endpoint: endpoint,
     apiKey: row.key,