package executor import ( "bufio" "bytes" "context" "fmt" "io" "net/http" "os" "path/filepath" "runtime" "strings" "time" kimiauth "github.com/router-for-me/CLIProxyAPI/v7/internal/auth/kimi" "github.com/router-for-me/CLIProxyAPI/v7/internal/config" "github.com/router-for-me/CLIProxyAPI/v7/internal/runtime/executor/helps" "github.com/router-for-me/CLIProxyAPI/v7/internal/thinking" "github.com/router-for-me/CLIProxyAPI/v7/internal/util" cliproxyauth "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/auth" cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/executor" sdktranslator "github.com/router-for-me/CLIProxyAPI/v7/sdk/translator" log "github.com/sirupsen/logrus" "github.com/tidwall/gjson" "github.com/tidwall/sjson") // KimiExecutor is a stateless executor for Kimi API using OpenAI-compatible chat completions.type KimiExecutor struct { ClaudeExecutor cfg *config.Config} // NewKimiExecutor creates a new Kimi executor.func NewKimiExecutor(cfg *config.Config) *KimiExecutor { return &KimiExecutor{cfg: cfg} } // Identifier returns the executor identifier.func (e *KimiExecutor) Identifier() string { return "kimi" } // PrepareRequest injects Kimi credentials into the outgoing HTTP request.func (e *KimiExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error { if req == nil { return nil } token := kimiCreds(auth) if strings.TrimSpace(token) != "" { req.Header.Set("Authorization", "Bearer "+token) } var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(req, attrs) return nil} // HttpRequest injects Kimi credentials into the request and executes it.func (e *KimiExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) { if req == nil { return nil, fmt.Errorf("kimi executor: request is nil") } if ctx == nil { ctx = req.Context() } httpReq := req.WithContext(ctx) if err := e.PrepareRequest(httpReq, auth); err != nil { return nil, err } httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) return httpClient.Do(httpReq)} // Execute performs a non-streaming chat completion request to Kimi.func (e *KimiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) { from := opts.SourceFormat if from.String() == "claude" { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.Execute(ctx, auth, req, opts) } baseModel := thinking.ParseSuffix(req.Model).ModelName token := kimiCreds(auth) reporter := helps.NewUsageReporter(ctx, e.Identifier(), baseModel, auth) defer reporter.TrackFailure(ctx, &err) to := sdktranslator.FromString("openai") originalPayloadSource := req.Payload if len(opts.OriginalRequest) > 0 { originalPayloadSource = opts.OriginalRequest } originalPayload := bytes.Clone(originalPayloadSource) originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false) body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false) // Strip kimi- prefix for upstream API upstreamModel := stripKimiPrefix(baseModel) body, err = sjson.SetBytes(body, "model", upstreamModel) if err != nil { return resp, fmt.Errorf("kimi executor: failed to set model in payload: %w", err) } body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier()) if err != nil { return resp, err } requestedModel := helps.PayloadRequestedModel(opts, req.Model) requestPath := helps.PayloadRequestPath(opts) body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers) body, err = normalizeKimiToolMessageLinks(body) if err != nil { return resp, err } url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions" httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body)) if err != nil { return resp, err } applyKimiHeadersWithAuth(httpReq, token, false, auth) var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(httpReq, attrs) var authID, authLabel, authType, authValue string if auth != nil { authID = auth.ID authLabel = auth.Label authType, authValue = auth.AccountInfo() } helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{ URL: url, Method: http.MethodPost, Headers: httpReq.Header.Clone(), Body: body, Provider: e.Identifier(), AuthID: authID, AuthLabel: authLabel, AuthType: authType, AuthValue: authValue, }) httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) httpResp, err := httpClient.Do(httpReq) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return resp, err } defer func() { if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } }() helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone()) if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 { b, _ := io.ReadAll(httpResp.Body) helps.AppendAPIResponseChunk(ctx, e.cfg, b) helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b)) err = statusErr{code: httpResp.StatusCode, msg: string(b)} return resp, err } data, err := io.ReadAll(httpResp.Body) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return resp, err } helps.AppendAPIResponseChunk(ctx, e.cfg, data) reporter.Publish(ctx, helps.ParseOpenAIUsage(data)) var param any // 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, ¶m) resp = cliproxyexecutor.Response{Payload: 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) (_ *cliproxyexecutor.StreamResult, err error) { from := opts.SourceFormat if from.String() == "claude" { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.ExecuteStream(ctx, auth, req, opts) } baseModel := thinking.ParseSuffix(req.Model).ModelName token := kimiCreds(auth) reporter := helps.NewUsageReporter(ctx, e.Identifier(), baseModel, auth) defer reporter.TrackFailure(ctx, &err) to := sdktranslator.FromString("openai") originalPayloadSource := req.Payload if len(opts.OriginalRequest) > 0 { originalPayloadSource = opts.OriginalRequest } originalPayload := bytes.Clone(originalPayloadSource) originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true) body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true) // Strip kimi- prefix for upstream API upstreamModel := stripKimiPrefix(baseModel) body, err = sjson.SetBytes(body, "model", upstreamModel) if err != nil { return nil, fmt.Errorf("kimi executor: failed to set model in payload: %w", err) } body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier()) if err != nil { return nil, err } body, err = sjson.SetBytes(body, "stream_options.include_usage", true) if err != nil { return nil, fmt.Errorf("kimi executor: failed to set stream_options in payload: %w", err) } requestedModel := helps.PayloadRequestedModel(opts, req.Model) requestPath := helps.PayloadRequestPath(opts) body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers) body, err = normalizeKimiToolMessageLinks(body) if err != nil { return nil, err } url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions" httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body)) if err != nil { return nil, err } applyKimiHeadersWithAuth(httpReq, token, true, auth) var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(httpReq, attrs) var authID, authLabel, authType, authValue string if auth != nil { authID = auth.ID authLabel = auth.Label authType, authValue = auth.AccountInfo() } helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{ URL: url, Method: http.MethodPost, Headers: httpReq.Header.Clone(), Body: body, Provider: e.Identifier(), AuthID: authID, AuthLabel: authLabel, AuthType: authType, AuthValue: authValue, }) httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) httpResp, err := httpClient.Do(httpReq) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return nil, err } helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone()) if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 { b, _ := io.ReadAll(httpResp.Body) helps.AppendAPIResponseChunk(ctx, e.cfg, b) helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b)) if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } err = statusErr{code: httpResp.StatusCode, msg: string(b)} return nil, err } out := make(chan cliproxyexecutor.StreamChunk) go func() { defer close(out) defer func() { if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } }() scanner := bufio.NewScanner(httpResp.Body) scanner.Buffer(nil, 1_048_576) // 1MB var param any for scanner.Scan() { line := scanner.Bytes() helps.AppendAPIResponseChunk(ctx, e.cfg, line) if detail, ok := helps.ParseOpenAIStreamUsage(line); ok { reporter.Publish(ctx, detail) } chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, bytes.Clone(line), ¶m) for i := range chunks { select { case out <- cliproxyexecutor.StreamChunk{Payload: chunks[i]}: case <-ctx.Done(): return } } } doneChunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, []byte("[DONE]"), ¶m) for i := range doneChunks { select { case out <- cliproxyexecutor.StreamChunk{Payload: doneChunks[i]}: case <-ctx.Done(): return } } if errScan := scanner.Err(); errScan != nil { helps.RecordAPIResponseError(ctx, e.cfg, errScan) reporter.PublishFailure(ctx, errScan) select { case out <- cliproxyexecutor.StreamChunk{Err: errScan}: case <-ctx.Done(): } } }() return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil} // CountTokens estimates token count for Kimi requests.func (e *KimiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.CountTokens(ctx, auth, req, opts)} func normalizeKimiToolMessageLinks(body []byte) ([]byte, error) { if len(body) == 0 || !gjson.ValidBytes(body) { return body, nil } messages := gjson.GetBytes(body, "messages") if !messages.Exists() || !messages.IsArray() { return body, nil } msgs := messages.Array() out, dropped, err := filterKimiEmptyAssistantMessages(body, msgs) if err != nil { return body, err } if dropped > 0 { log.WithField("dropped_assistant_messages", dropped).Debug("kimi executor: dropped empty assistant messages") } messages = gjson.GetBytes(out, "messages") msgs = messages.Array() pending := make([]string, 0) patched := 0 patchedReasoning := 0 ambiguous := 0 latestReasoning := "" hasLatestReasoning := false removePending := func(id string) { for idx := range pending { if pending[idx] != id { continue } pending = append(pending[:idx], pending[idx+1:]...) return } } for msgIdx := range msgs { msg := msgs[msgIdx] role := strings.TrimSpace(msg.Get("role").String()) switch role { case "assistant": reasoning := msg.Get("reasoning_content") if reasoning.Exists() { reasoningText := reasoning.String() if strings.TrimSpace(reasoningText) != "" { latestReasoning = reasoningText hasLatestReasoning = true } } toolCalls := msg.Get("tool_calls") if !toolCalls.Exists() || !toolCalls.IsArray() || len(toolCalls.Array()) == 0 { continue } if !reasoning.Exists() || strings.TrimSpace(reasoning.String()) == "" { reasoningText := fallbackAssistantReasoning(msg, hasLatestReasoning, latestReasoning) path := fmt.Sprintf("messages.%d.reasoning_content", msgIdx) next, err := sjson.SetBytes(out, path, reasoningText) if err != nil { return body, fmt.Errorf("kimi executor: failed to set assistant reasoning_content: %w", err) } out = next patchedReasoning++ } for _, tc := range toolCalls.Array() { id := strings.TrimSpace(tc.Get("id").String()) if id == "" { continue } pending = append(pending, id) } case "tool": toolCallID := strings.TrimSpace(msg.Get("tool_call_id").String()) if toolCallID == "" { toolCallID = strings.TrimSpace(msg.Get("call_id").String()) if toolCallID != "" { path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx) next, err := sjson.SetBytes(out, path, toolCallID) if err != nil { return body, fmt.Errorf("kimi executor: failed to set tool_call_id from call_id: %w", err) } out = next patched++ } } if toolCallID == "" { if len(pending) == 1 { toolCallID = pending[0] path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx) next, err := sjson.SetBytes(out, path, toolCallID) if err != nil { return body, fmt.Errorf("kimi executor: failed to infer tool_call_id: %w", err) } out = next patched++ } else if len(pending) > 1 { ambiguous++ } } if toolCallID != "" { removePending(toolCallID) } } } if patched > 0 || patchedReasoning > 0 { log.WithFields(log.Fields{ "patched_tool_messages": patched, "patched_reasoning_messages": patchedReasoning, }).Debug("kimi executor: normalized tool message fields") } if ambiguous > 0 { log.WithFields(log.Fields{ "ambiguous_tool_messages": ambiguous, "pending_tool_calls": len(pending), }).Warn("kimi executor: tool messages missing tool_call_id with ambiguous candidates") } return out, nil} func filterKimiEmptyAssistantMessages(body []byte, msgs []gjson.Result) ([]byte, int, error) { kept := make([]string, 0, len(msgs)) dropped := 0 for _, msg := range msgs { if shouldDropKimiAssistantMessage(msg) { dropped++ continue } kept = append(kept, msg.Raw) } if dropped == 0 { return body, 0, nil } rawMessages := []byte("[" + strings.Join(kept, ",") + "]") out, err := sjson.SetRawBytes(body, "messages", rawMessages) if err != nil { return body, 0, fmt.Errorf("kimi executor: failed to drop empty assistant messages: %w", err) } return out, dropped, nil} func shouldDropKimiAssistantMessage(msg gjson.Result) bool { if strings.TrimSpace(msg.Get("role").String()) != "assistant" { return false } if hasKimiToolCalls(msg) || hasKimiLegacyFunctionCall(msg) || hasKimiAssistantReasoning(msg) { return false } return isKimiAssistantContentEmpty(msg.Get("content"))} func hasKimiToolCalls(msg gjson.Result) bool { toolCalls := msg.Get("tool_calls") return toolCalls.Exists() && toolCalls.IsArray() && len(toolCalls.Array()) > 0} func hasKimiLegacyFunctionCall(msg gjson.Result) bool { functionCall := msg.Get("function_call") if !functionCall.Exists() || functionCall.Type == gjson.Null { return false } if functionCall.IsObject() && strings.TrimSpace(functionCall.Raw) == "{}" { return false } return strings.TrimSpace(functionCall.Raw) != ""} func hasKimiAssistantReasoning(msg gjson.Result) bool { reasoning := msg.Get("reasoning_content")Lines 1–500 of 749. Load more
package executor import ( "bufio" "bytes" "context" "fmt" "io" "net/http" "os" "path/filepath" "runtime" "strings" "time" kimiauth "github.com/router-for-me/CLIProxyAPI/v7/internal/auth/kimi" "github.com/router-for-me/CLIProxyAPI/v7/internal/config" "github.com/router-for-me/CLIProxyAPI/v7/internal/runtime/executor/helps" "github.com/router-for-me/CLIProxyAPI/v7/internal/thinking" "github.com/router-for-me/CLIProxyAPI/v7/internal/util" cliproxyauth "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/auth" cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/executor" sdktranslator "github.com/router-for-me/CLIProxyAPI/v7/sdk/translator" log "github.com/sirupsen/logrus" "github.com/tidwall/gjson" "github.com/tidwall/sjson" ) // KimiExecutor is a stateless executor for Kimi API using OpenAI-compatible chat completions. type KimiExecutor struct { ClaudeExecutor cfg *config.Config } // NewKimiExecutor creates a new Kimi executor. func NewKimiExecutor(cfg *config.Config) *KimiExecutor { return &KimiExecutor{cfg: cfg} } // Identifier returns the executor identifier. func (e *KimiExecutor) Identifier() string { return "kimi" } // PrepareRequest injects Kimi credentials into the outgoing HTTP request. func (e *KimiExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error { if req == nil { return nil } token := kimiCreds(auth) if strings.TrimSpace(token) != "" { req.Header.Set("Authorization", "Bearer "+token) } var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(req, attrs) return nil } // HttpRequest injects Kimi credentials into the request and executes it. func (e *KimiExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) { if req == nil { return nil, fmt.Errorf("kimi executor: request is nil") } if ctx == nil { ctx = req.Context() } httpReq := req.WithContext(ctx) if err := e.PrepareRequest(httpReq, auth); err != nil { return nil, err } httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) return httpClient.Do(httpReq) } // Execute performs a non-streaming chat completion request to Kimi. func (e *KimiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) { from := opts.SourceFormat if from.String() == "claude" { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.Execute(ctx, auth, req, opts) } baseModel := thinking.ParseSuffix(req.Model).ModelName token := kimiCreds(auth) reporter := helps.NewUsageReporter(ctx, e.Identifier(), baseModel, auth) defer reporter.TrackFailure(ctx, &err) to := sdktranslator.FromString("openai") originalPayloadSource := req.Payload if len(opts.OriginalRequest) > 0 { originalPayloadSource = opts.OriginalRequest } originalPayload := bytes.Clone(originalPayloadSource) originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false) body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false) // Strip kimi- prefix for upstream API upstreamModel := stripKimiPrefix(baseModel) body, err = sjson.SetBytes(body, "model", upstreamModel) if err != nil { return resp, fmt.Errorf("kimi executor: failed to set model in payload: %w", err) } body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier()) if err != nil { return resp, err } requestedModel := helps.PayloadRequestedModel(opts, req.Model) requestPath := helps.PayloadRequestPath(opts) body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers) body, err = normalizeKimiToolMessageLinks(body) if err != nil { return resp, err } url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions" httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body)) if err != nil { return resp, err } applyKimiHeadersWithAuth(httpReq, token, false, auth) var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(httpReq, attrs) var authID, authLabel, authType, authValue string if auth != nil { authID = auth.ID authLabel = auth.Label authType, authValue = auth.AccountInfo() } helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{ URL: url, Method: http.MethodPost, Headers: httpReq.Header.Clone(), Body: body, Provider: e.Identifier(), AuthID: authID, AuthLabel: authLabel, AuthType: authType, AuthValue: authValue, }) httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) httpResp, err := httpClient.Do(httpReq) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return resp, err } defer func() { if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } }() helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone()) if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 { b, _ := io.ReadAll(httpResp.Body) helps.AppendAPIResponseChunk(ctx, e.cfg, b) helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b)) err = statusErr{code: httpResp.StatusCode, msg: string(b)} return resp, err } data, err := io.ReadAll(httpResp.Body) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return resp, err } helps.AppendAPIResponseChunk(ctx, e.cfg, data) reporter.Publish(ctx, helps.ParseOpenAIUsage(data)) var param any // 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, ¶m) resp = cliproxyexecutor.Response{Payload: 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) (_ *cliproxyexecutor.StreamResult, err error) { from := opts.SourceFormat if from.String() == "claude" { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.ExecuteStream(ctx, auth, req, opts) } baseModel := thinking.ParseSuffix(req.Model).ModelName token := kimiCreds(auth) reporter := helps.NewUsageReporter(ctx, e.Identifier(), baseModel, auth) defer reporter.TrackFailure(ctx, &err) to := sdktranslator.FromString("openai") originalPayloadSource := req.Payload if len(opts.OriginalRequest) > 0 { originalPayloadSource = opts.OriginalRequest } originalPayload := bytes.Clone(originalPayloadSource) originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true) body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true) // Strip kimi- prefix for upstream API upstreamModel := stripKimiPrefix(baseModel) body, err = sjson.SetBytes(body, "model", upstreamModel) if err != nil { return nil, fmt.Errorf("kimi executor: failed to set model in payload: %w", err) } body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier()) if err != nil { return nil, err } body, err = sjson.SetBytes(body, "stream_options.include_usage", true) if err != nil { return nil, fmt.Errorf("kimi executor: failed to set stream_options in payload: %w", err) } requestedModel := helps.PayloadRequestedModel(opts, req.Model) requestPath := helps.PayloadRequestPath(opts) body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers) body, err = normalizeKimiToolMessageLinks(body) if err != nil { return nil, err } url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions" httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body)) if err != nil { return nil, err } applyKimiHeadersWithAuth(httpReq, token, true, auth) var attrs map[string]string if auth != nil { attrs = auth.Attributes } util.ApplyCustomHeadersFromAttrs(httpReq, attrs) var authID, authLabel, authType, authValue string if auth != nil { authID = auth.ID authLabel = auth.Label authType, authValue = auth.AccountInfo() } helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{ URL: url, Method: http.MethodPost, Headers: httpReq.Header.Clone(), Body: body, Provider: e.Identifier(), AuthID: authID, AuthLabel: authLabel, AuthType: authType, AuthValue: authValue, }) httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0) httpResp, err := httpClient.Do(httpReq) if err != nil { helps.RecordAPIResponseError(ctx, e.cfg, err) return nil, err } helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone()) if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 { b, _ := io.ReadAll(httpResp.Body) helps.AppendAPIResponseChunk(ctx, e.cfg, b) helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b)) if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } err = statusErr{code: httpResp.StatusCode, msg: string(b)} return nil, err } out := make(chan cliproxyexecutor.StreamChunk) go func() { defer close(out) defer func() { if errClose := httpResp.Body.Close(); errClose != nil { log.Errorf("kimi executor: close response body error: %v", errClose) } }() scanner := bufio.NewScanner(httpResp.Body) scanner.Buffer(nil, 1_048_576) // 1MB var param any for scanner.Scan() { line := scanner.Bytes() helps.AppendAPIResponseChunk(ctx, e.cfg, line) if detail, ok := helps.ParseOpenAIStreamUsage(line); ok { reporter.Publish(ctx, detail) } chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, bytes.Clone(line), ¶m) for i := range chunks { select { case out <- cliproxyexecutor.StreamChunk{Payload: chunks[i]}: case <-ctx.Done(): return } } } doneChunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, []byte("[DONE]"), ¶m) for i := range doneChunks { select { case out <- cliproxyexecutor.StreamChunk{Payload: doneChunks[i]}: case <-ctx.Done(): return } } if errScan := scanner.Err(); errScan != nil { helps.RecordAPIResponseError(ctx, e.cfg, errScan) reporter.PublishFailure(ctx, errScan) select { case out <- cliproxyexecutor.StreamChunk{Err: errScan}: case <-ctx.Done(): } } }() return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil } // CountTokens estimates token count for Kimi requests. func (e *KimiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) { auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL return e.ClaudeExecutor.CountTokens(ctx, auth, req, opts) } func normalizeKimiToolMessageLinks(body []byte) ([]byte, error) { if len(body) == 0 || !gjson.ValidBytes(body) { return body, nil } messages := gjson.GetBytes(body, "messages") if !messages.Exists() || !messages.IsArray() { return body, nil } msgs := messages.Array() out, dropped, err := filterKimiEmptyAssistantMessages(body, msgs) if err != nil { return body, err } if dropped > 0 { log.WithField("dropped_assistant_messages", dropped).Debug("kimi executor: dropped empty assistant messages") } messages = gjson.GetBytes(out, "messages") msgs = messages.Array() pending := make([]string, 0) patched := 0 patchedReasoning := 0 ambiguous := 0 latestReasoning := "" hasLatestReasoning := false removePending := func(id string) { for idx := range pending { if pending[idx] != id { continue } pending = append(pending[:idx], pending[idx+1:]...) return } } for msgIdx := range msgs { msg := msgs[msgIdx] role := strings.TrimSpace(msg.Get("role").String()) switch role { case "assistant": reasoning := msg.Get("reasoning_content") if reasoning.Exists() { reasoningText := reasoning.String() if strings.TrimSpace(reasoningText) != "" { latestReasoning = reasoningText hasLatestReasoning = true } } toolCalls := msg.Get("tool_calls") if !toolCalls.Exists() || !toolCalls.IsArray() || len(toolCalls.Array()) == 0 { continue } if !reasoning.Exists() || strings.TrimSpace(reasoning.String()) == "" { reasoningText := fallbackAssistantReasoning(msg, hasLatestReasoning, latestReasoning) path := fmt.Sprintf("messages.%d.reasoning_content", msgIdx) next, err := sjson.SetBytes(out, path, reasoningText) if err != nil { return body, fmt.Errorf("kimi executor: failed to set assistant reasoning_content: %w", err) } out = next patchedReasoning++ } for _, tc := range toolCalls.Array() { id := strings.TrimSpace(tc.Get("id").String()) if id == "" { continue } pending = append(pending, id) } case "tool": toolCallID := strings.TrimSpace(msg.Get("tool_call_id").String()) if toolCallID == "" { toolCallID = strings.TrimSpace(msg.Get("call_id").String()) if toolCallID != "" { path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx) next, err := sjson.SetBytes(out, path, toolCallID) if err != nil { return body, fmt.Errorf("kimi executor: failed to set tool_call_id from call_id: %w", err) } out = next patched++ } } if toolCallID == "" { if len(pending) == 1 { toolCallID = pending[0] path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx) next, err := sjson.SetBytes(out, path, toolCallID) if err != nil { return body, fmt.Errorf("kimi executor: failed to infer tool_call_id: %w", err) } out = next patched++ } else if len(pending) > 1 { ambiguous++ } } if toolCallID != "" { removePending(toolCallID) } } } if patched > 0 || patchedReasoning > 0 { log.WithFields(log.Fields{ "patched_tool_messages": patched, "patched_reasoning_messages": patchedReasoning, }).Debug("kimi executor: normalized tool message fields") } if ambiguous > 0 { log.WithFields(log.Fields{ "ambiguous_tool_messages": ambiguous, "pending_tool_calls": len(pending), }).Warn("kimi executor: tool messages missing tool_call_id with ambiguous candidates") } return out, nil } func filterKimiEmptyAssistantMessages(body []byte, msgs []gjson.Result) ([]byte, int, error) { kept := make([]string, 0, len(msgs)) dropped := 0 for _, msg := range msgs { if shouldDropKimiAssistantMessage(msg) { dropped++ continue } kept = append(kept, msg.Raw) } if dropped == 0 { return body, 0, nil } rawMessages := []byte("[" + strings.Join(kept, ",") + "]") out, err := sjson.SetRawBytes(body, "messages", rawMessages) if err != nil { return body, 0, fmt.Errorf("kimi executor: failed to drop empty assistant messages: %w", err) } return out, dropped, nil } func shouldDropKimiAssistantMessage(msg gjson.Result) bool { if strings.TrimSpace(msg.Get("role").String()) != "assistant" { return false } if hasKimiToolCalls(msg) || hasKimiLegacyFunctionCall(msg) || hasKimiAssistantReasoning(msg) { return false } return isKimiAssistantContentEmpty(msg.Get("content")) } func hasKimiToolCalls(msg gjson.Result) bool { toolCalls := msg.Get("tool_calls") return toolCalls.Exists() && toolCalls.IsArray() && len(toolCalls.Array()) > 0 } func hasKimiLegacyFunctionCall(msg gjson.Result) bool { functionCall := msg.Get("function_call") if !functionCall.Exists() || functionCall.Type == gjson.Null { return false } if functionCall.IsObject() && strings.TrimSpace(functionCall.Raw) == "{}" { return false } return strings.TrimSpace(functionCall.Raw) != "" } func hasKimiAssistantReasoning(msg gjson.Result) bool { reasoning := msg.Get("reasoning_content")