chuan/CLIProxyAPI
1
0
Fork 0
mirror of https://github.com/router-for-me/CLIProxyAPI.git synced 2026-02-28 15:25:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: passthrough upstream response headers to clients

Browse Source 
CPA previously stripped ALL response headers from upstream AI provider
APIs, preventing clients from seeing rate-limit info, request IDs,
server-timing and other useful headers.

Changes:
- Add Headers field to Response and StreamResult structs
- Add FilterUpstreamHeaders helper (hop-by-hop + security denylist)
- Add WriteUpstreamHeaders helper (respects CPA-set headers)
- ExecuteWithAuthManager/ExecuteCountWithAuthManager now return headers
- ExecuteStreamWithAuthManager returns headers from initial connection
- All 11 provider executors populate Response.Headers
- All handler call sites write filtered upstream headers before response

Filtered headers (not forwarded):
- RFC 7230 hop-by-hop: Connection, Transfer-Encoding, Keep-Alive, etc.
- Security: Set-Cookie
- CPA-managed: Content-Length, Content-Encoding
This commit is contained in:
Kirill Turanskiy
2026-02-18 00:16:22 +03:00
parent 453aaf8774
commit 1f8f198c45
24 changed files with 192 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
internal/runtime/executor/aistudio_executor.go
View File

@@ -164,12 +164,12 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
reporter.publish(ctx, parseGeminiUsage(wsResp.Body))
var param any
out := sdktranslator.TranslateNonStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, wsResp.Body, &param)
resp = cliproxyexecutor.Response{Payload: ensureColonSpacedJSON([]byte(out))}
resp = cliproxyexecutor.Response{Payload: ensureColonSpacedJSON([]byte(out)), Headers: wsResp.Headers.Clone()}
return resp, nil
}
// ExecuteStream performs a streaming request to the AI Studio API.
func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -254,7 +254,6 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
return nil, statusErr{code: firstEvent.Status, msg: body.String()}
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func(first wsrelay.StreamEvent) {
defer close(out)
var param any
@@ -318,7 +317,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
}
}
}(firstEvent)
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: firstEvent.Headers.Clone(), Chunks: out}, nil
}
// CountTokens counts tokens for the given request using the AI Studio API.

11
internal/runtime/executor/antigravity_executor.go
View File

@@ -232,7 +232,7 @@ attemptLoop:
reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
var param any
converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, bodyBytes, &param)
resp = cliproxyexecutor.Response{Payload: []byte(converted)}
resp = cliproxyexecutor.Response{Payload: []byte(converted), Headers: httpResp.Header.Clone()}
reporter.ensurePublished(ctx)
return resp, nil
}
@@ -436,7 +436,7 @@ attemptLoop:
reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
var param any
converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, resp.Payload, &param)
resp = cliproxyexecutor.Response{Payload: []byte(converted)}
resp = cliproxyexecutor.Response{Payload: []byte(converted), Headers: httpResp.Header.Clone()}
reporter.ensurePublished(ctx)
return resp, nil
@@ -645,7 +645,7 @@ func (e *AntigravityExecutor) convertStreamToNonStream(stream []byte) []byte {
}
// ExecuteStream performs a streaming request to the Antigravity API.
func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -775,7 +775,6 @@ attemptLoop:
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func(resp *http.Response) {
defer close(out)
defer func() {
@@ -820,7 +819,7 @@ attemptLoop:
reporter.ensurePublished(ctx)
}
}(httpResp)
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
switch {
@@ -968,7 +967,7 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
if httpResp.StatusCode >= http.StatusOK && httpResp.StatusCode < http.StatusMultipleChoices {
count := gjson.GetBytes(bodyBytes, "totalTokens").Int()
translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, bodyBytes)
return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
return cliproxyexecutor.Response{Payload: []byte(translated), Headers: httpResp.Header.Clone()}, nil
}
lastStatus = httpResp.StatusCode

9
internal/runtime/executor/claude_executor.go
View File

@@ -222,11 +222,11 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
data,
&param,
)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -329,7 +329,6 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -398,7 +397,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
@@ -487,7 +486,7 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
appendAPIResponseChunk(ctx, e.cfg, data)
count := gjson.GetBytes(data, "input_tokens").Int()
out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
return cliproxyexecutor.Response{Payload: []byte(out)}, nil
return cliproxyexecutor.Response{Payload: []byte(out), Headers: resp.Header.Clone()}, nil
}
func (e *ClaudeExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {

9
internal/runtime/executor/codex_executor.go
View File

@@ -183,7 +183,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, originalPayload, body, line, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
err = statusErr{code: 408, msg: "stream error: stream disconnected before completion: stream closed before response.completed"}
@@ -273,11 +273,11 @@ func (e *CodexExecutor) executeCompact(ctx context.Context, auth *cliproxyauth.A
reporter.ensurePublished(ctx)
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, originalPayload, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusBadRequest, msg: "streaming not supported for /responses/compact"}
}
@@ -362,7 +362,6 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -397,7 +396,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {

9
internal/runtime/executor/gemini_cli_executor.go
View File

@@ -225,7 +225,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
reporter.publish(ctx, parseGeminiCLIUsage(data))
var param any
out := sdktranslator.TranslateNonStream(respCtx, to, from, attemptModel, opts.OriginalRequest, payload, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
@@ -256,7 +256,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
}
// ExecuteStream performs a streaming request to the Gemini CLI API.
func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -382,7 +382,6 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func(resp *http.Response, reqBody []byte, attemptModel string) {
defer close(out)
defer func() {
@@ -441,7 +440,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
}
}(httpResp, append([]byte(nil), payload...), attemptModel)
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
if len(lastBody) > 0 {
@@ -546,7 +545,7 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
if resp.StatusCode >= 200 && resp.StatusCode < 300 {
count := gjson.GetBytes(data, "totalTokens").Int()
translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, data)
return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
return cliproxyexecutor.Response{Payload: []byte(translated), Headers: resp.Header.Clone()}, nil
}
lastStatus = resp.StatusCode
lastBody = append([]byte(nil), data...)

9
internal/runtime/executor/gemini_executor.go
View File

@@ -205,12 +205,12 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
reporter.publish(ctx, parseGeminiUsage(data))
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
// ExecuteStream performs a streaming request to the Gemini API.
func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -298,7 +298,6 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -335,7 +334,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// CountTokens counts tokens for the given request using the Gemini API.
@@ -416,7 +415,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
count := gjson.GetBytes(data, "totalTokens").Int()
translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, data)
return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
return cliproxyexecutor.Response{Payload: []byte(translated), Headers: resp.Header.Clone()}, nil
}
// Refresh refreshes the authentication credentials (no-op for Gemini API key).

20
internal/runtime/executor/gemini_vertex_executor.go
View File

@@ -253,7 +253,7 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
}
// ExecuteStream performs a streaming request to the Vertex AI API.
func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -419,7 +419,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
to := sdktranslator.FromString("gemini")
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
@@ -524,12 +524,12 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
reporter.publish(ctx, parseGeminiUsage(data))
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
// executeStreamWithServiceAccount handles streaming authentication using service account credentials.
func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (_ *cliproxyexecutor.StreamResult, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -618,7 +618,6 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -650,11 +649,11 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// executeStreamWithAPIKey handles streaming authentication using API key credentials.
func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (_ *cliproxyexecutor.StreamResult, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -743,7 +742,6 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -775,7 +773,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// countTokensWithServiceAccount counts tokens using service account credentials.
@@ -859,7 +857,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
appendAPIResponseChunk(ctx, e.cfg, data)
count := gjson.GetBytes(data, "totalTokens").Int()
out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
return cliproxyexecutor.Response{Payload: []byte(out)}, nil
return cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}, nil
}
// countTokensWithAPIKey handles token counting using API key credentials.
@@ -943,7 +941,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
appendAPIResponseChunk(ctx, e.cfg, data)
count := gjson.GetBytes(data, "totalTokens").Int()
out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
return cliproxyexecutor.Response{Payload: []byte(out)}, nil
return cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}, nil
}
// vertexCreds extracts project, location and raw service account JSON from auth metadata.

7
internal/runtime/executor/iflow_executor.go
View File

@@ -169,12 +169,12 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
// the original model name in the response for client compatibility.
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
// ExecuteStream performs a streaming chat completion request.
func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -262,7 +262,6 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -294,7 +293,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
reporter.ensurePublished(ctx)
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
func (e *IFlowExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {

7
internal/runtime/executor/kimi_executor.go
View File

@@ -161,12 +161,12 @@ func (e *KimiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
// the original model name in the response for client compatibility.
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
// ExecuteStream performs a streaming chat completion request to Kimi.
func (e *KimiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *KimiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
from := opts.SourceFormat
if from.String() == "claude" {
auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL
@@ -253,7 +253,6 @@ func (e *KimiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -285,7 +284,7 @@ func (e *KimiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// CountTokens estimates token count for Kimi requests.

7
internal/runtime/executor/openai_compat_executor.go
View File

@@ -172,11 +172,11 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
// Translate response back to source format when needed
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, body, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -258,7 +258,6 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -298,7 +297,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
// Ensure we record the request if no usage chunk was ever seen
reporter.ensurePublished(ctx)
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
func (e *OpenAICompatExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {

7
internal/runtime/executor/qwen_executor.go
View File

@@ -150,11 +150,11 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
// the original model name in the response for client compatibility.
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
return resp, nil
}
func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
@@ -236,7 +236,6 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
@@ -268,7 +267,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
func (e *QwenExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {