diff --git a/README.md b/README.md index f3b518f..2f6f25b 100644 --- a/README.md +++ b/README.md @@ -108,6 +108,8 @@ clawgo provider login codex --manual - 额度或限流失败时自动切到 OAuth 账号池 - 仍保留多账号轮换和后台刷新 +如果某个 OpenAI 兼容服务商只支持 `POST /v1/chat/completions`，可以在对应 provider 配置里设置 `responses.api: "chat_completions"`；默认值是 `responses`。 + ### 4. 启动 交互模式： diff --git a/README_EN.md b/README_EN.md index a8babbd..9cf2917 100644 --- a/README_EN.md +++ b/README_EN.md @@ -119,6 +119,8 @@ If you have both an `API key` and OAuth accounts for the same upstream, prefer c - the provider runtime panel shows current candidate ordering, the most recent successful credential, and recent hit/error history - to persist runtime history across restarts, configure `runtime_persist`, `runtime_history_file`, and `runtime_history_max` on the provider +If an OpenAI-compatible provider only supports `POST /v1/chat/completions`, set `responses.api: "chat_completions"` on that provider. The default remains `responses`. + ### 4. Start Interactive mode: diff --git a/config.example.json b/config.example.json index 9ec4bf8..6cde3b1 100644 --- a/config.example.json +++ b/config.example.json @@ -180,6 +180,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], @@ -208,6 +209,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], @@ -237,6 +239,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], @@ -253,6 +256,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], @@ -280,6 +284,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], @@ -306,6 +311,7 @@ "max_tokens": 8192, "temperature": 0.7, "responses": { + "api": "responses", "web_search_enabled": false, "web_search_context_size": "", "file_search_vector_store_ids": [], diff --git a/pkg/config/config.go b/pkg/config/config.go index 61e26f4..a02486e 100644 --- a/pkg/config/config.go +++ b/pkg/config/config.go @@ -260,6 +260,7 @@ type ProviderOAuthConfig struct { } type ProviderResponsesConfig struct { + API string `json:"api,omitempty"` WebSearchEnabled bool `json:"web_search_enabled"` WebSearchContextSize string `json:"web_search_context_size"` FileSearchVectorStoreIDs []string `json:"file_search_vector_store_ids"` diff --git a/pkg/config/normalized_test.go b/pkg/config/normalized_test.go index 520e08b..b8947d5 100644 --- a/pkg/config/normalized_test.go +++ b/pkg/config/normalized_test.go @@ -11,6 +11,9 @@ func TestNormalizedViewProjectsCoreAndRuntime(t *testing.T) { MaxTokens: 12288, Temperature: 0.35, TimeoutSec: 90, + Responses: ProviderResponsesConfig{ + API: "chat_completions", + }, } cfg.Agents.Subagents["coder"] = SubagentConfig{ Enabled: true, @@ -40,4 +43,7 @@ func TestNormalizedViewProjectsCoreAndRuntime(t *testing.T) { if got := view.Runtime.Providers["openai"].Temperature; got != 0.35 { t.Fatalf("expected provider temperature in normalized runtime view, got %v", got) } + if got := view.Runtime.Providers["openai"].Responses.API; got != "chat_completions" { + t.Fatalf("expected provider responses.api in normalized runtime view, got %q", got) + } } diff --git a/pkg/config/validate.go b/pkg/config/validate.go index ac2886f..56fe052 100644 --- a/pkg/config/validate.go +++ b/pkg/config/validate.go @@ -515,6 +515,13 @@ func validateProviderConfig(path string, p ProviderConfig) []error { if p.OAuth.CooldownSec < 0 { errs = append(errs, fmt.Errorf("%s.oauth.cooldown_sec must be >= 0", path)) } + if p.Responses.API != "" { + switch strings.TrimSpace(p.Responses.API) { + case "responses", "chat_completions": + default: + errs = append(errs, fmt.Errorf("%s.responses.api must be one of: responses, chat_completions", path)) + } + } if p.Responses.WebSearchContextSize != "" { switch p.Responses.WebSearchContextSize { case "low", "medium", "high": diff --git a/pkg/config/validate_test.go b/pkg/config/validate_test.go index e1df981..3567fe1 100644 --- a/pkg/config/validate_test.go +++ b/pkg/config/validate_test.go @@ -247,3 +247,27 @@ func TestValidateProviderHybridRequiresOAuthProvider(t *testing.T) { t.Fatalf("expected oauth.provider validation error, got %v", errs) } } + +func TestValidateProviderResponsesAPIRejectsUnknownValue(t *testing.T) { + t.Parallel() + + cfg := DefaultConfig() + pc := cfg.Models.Providers["openai"] + pc.Responses.API = "legacy" + cfg.Models.Providers["openai"] = pc + + errs := Validate(cfg) + if len(errs) == 0 { + t.Fatalf("expected validation errors") + } + found := false + for _, err := range errs { + if strings.Contains(err.Error(), "models.providers.openai.responses.api") { + found = true + break + } + } + if !found { + t.Fatalf("expected responses.api validation error, got %v", errs) + } +} diff --git a/pkg/providers/http_provider.go b/pkg/providers/http_provider.go index 38c68e6..e10688c 100644 --- a/pkg/providers/http_provider.go +++ b/pkg/providers/http_provider.go @@ -31,6 +31,7 @@ type HTTPProvider struct { apiBase string defaultModel string supportsResponsesCompact bool + responsesAPI string authMode string timeout time.Duration httpClient *http.Client @@ -48,6 +49,7 @@ func NewHTTPProvider(providerName, apiKey, apiBase, defaultModel string, support apiBase: normalizedBase, defaultModel: strings.TrimSpace(defaultModel), supportsResponsesCompact: supportsResponsesCompact, + responsesAPI: "responses", authMode: authMode, timeout: timeout, httpClient: &http.Client{Timeout: timeout}, @@ -79,7 +81,7 @@ func (p *HTTPProvider) Chat(ctx context.Context, messages []Message, tools []Too if !json.Valid(body) { return nil, fmt.Errorf("API error (status %d, content-type %q): non-JSON response: %s", statusCode, contentType, previewResponseBody(body)) } - if p.useOpenAICompatChatUpstream() { + if p.useOpenAICompatChatUpstream() || p.useConfiguredOpenAICompatChat() { return parseOpenAICompatResponse(body) } return parseResponsesAPIResponse(body) @@ -102,7 +104,7 @@ func (p *HTTPProvider) ChatStream(ctx context.Context, messages []Message, tools if !json.Valid(body) { return nil, fmt.Errorf("API error (status %d, content-type %q): non-JSON response: %s", status, ctype, previewResponseBody(body)) } - if p.useOpenAICompatChatUpstream() { + if p.useOpenAICompatChatUpstream() || p.useConfiguredOpenAICompatChat() { return parseOpenAICompatResponse(body) } return parseResponsesAPIResponse(body) diff --git a/pkg/providers/openai_compat_adapter.go b/pkg/providers/openai_compat_adapter.go index f0376a2..cdb7b63 100644 --- a/pkg/providers/openai_compat_adapter.go +++ b/pkg/providers/openai_compat_adapter.go @@ -112,6 +112,18 @@ func (p *HTTPProvider) useOpenAICompatChatUpstream() bool { } } +func (p *HTTPProvider) useConfiguredOpenAICompatChat() bool { + if p == nil { + return false + } + switch strings.ToLower(strings.TrimSpace(p.responsesAPI)) { + case "chat_completions": + return true + default: + return false + } +} + func (p *HTTPProvider) compatBase() string { switch p.oauthProvider() { case defaultQwenOAuthProvider: diff --git a/pkg/providers/openai_compat_provider_test.go b/pkg/providers/openai_compat_provider_test.go index 4a72a2d..cce6186 100644 --- a/pkg/providers/openai_compat_provider_test.go +++ b/pkg/providers/openai_compat_provider_test.go @@ -1,6 +1,7 @@ package providers import ( + "encoding/json" "net/http" "net/http/httptest" "strings" @@ -180,3 +181,37 @@ func TestBuildOpenAICompatChatRequestStripsKimiPrefixAndSuffix(t *testing.T) { t.Fatalf("reasoning_effort = %#v, want auto", got) } } + +func TestHTTPProviderChatUsesConfiguredChatCompletionsAPI(t *testing.T) { + var gotPath string + var gotBody map[string]interface{} + server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + gotPath = r.URL.Path + if err := json.NewDecoder(r.Body).Decode(&gotBody); err != nil { + t.Fatalf("decode request: %v", err) + } + w.Header().Set("Content-Type", "application/json") + _, _ = w.Write([]byte(`{"choices":[{"message":{"content":"hello from chat"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":2,"total_tokens":3}}`)) + })) + defer server.Close() + + provider := NewHTTPProvider("openai", "token", server.URL+"/v1", "gpt-5", false, "api_key", 5*time.Second, nil) + provider.responsesAPI = "chat_completions" + + resp, err := provider.Chat(t.Context(), []Message{{Role: "user", Content: "hi"}}, nil, "gpt-5", nil) + if err != nil { + t.Fatalf("Chat error: %v", err) + } + if gotPath != "/v1/chat/completions" { + t.Fatalf("path = %q, want /v1/chat/completions", gotPath) + } + if gotBody["model"] != "gpt-5" { + t.Fatalf("model = %#v, want gpt-5", gotBody["model"]) + } + if resp.Content != "hello from chat" { + t.Fatalf("content = %q, want hello from chat", resp.Content) + } + if resp.Usage == nil || resp.Usage.TotalTokens != 3 { + t.Fatalf("usage = %#v, want total_tokens=3", resp.Usage) + } +} diff --git a/pkg/providers/provider_registry.go b/pkg/providers/provider_registry.go index 8a572ee..160cdfa 100644 --- a/pkg/providers/provider_registry.go +++ b/pkg/providers/provider_registry.go @@ -116,7 +116,11 @@ func CreateProviderByName(cfg *config.Config, name string) (LLMProvider, error) if oauthProvider == defaultIFlowOAuthProvider || strings.EqualFold(routeName, defaultIFlowOAuthProvider) { return NewIFlowProvider(routeName, pc.APIKey, pc.APIBase, defaultModel, pc.SupportsResponsesCompact, pc.Auth, time.Duration(pc.TimeoutSec)*time.Second, oauth), nil } - return NewHTTPProvider(routeName, pc.APIKey, pc.APIBase, defaultModel, pc.SupportsResponsesCompact, pc.Auth, time.Duration(pc.TimeoutSec)*time.Second, oauth), nil + provider := NewHTTPProvider(routeName, pc.APIKey, pc.APIBase, defaultModel, pc.SupportsResponsesCompact, pc.Auth, time.Duration(pc.TimeoutSec)*time.Second, oauth) + if api := strings.TrimSpace(pc.Responses.API); api != "" { + provider.responsesAPI = api + } + return provider, nil } func ProviderSupportsResponsesCompact(cfg *config.Config, name string) bool { diff --git a/pkg/providers/responses_adapter.go b/pkg/providers/responses_adapter.go index 2037b21..874ed5a 100644 --- a/pkg/providers/responses_adapter.go +++ b/pkg/providers/responses_adapter.go @@ -44,7 +44,7 @@ func (p *HTTPProvider) callResponses(ctx context.Context, messages []Message, to if prevID, ok := stringOption(options, "responses_previous_response_id"); ok && prevID != "" { requestBody["previous_response_id"] = prevID } - if p.useOpenAICompatChatUpstream() { + if p.useOpenAICompatChatUpstream() || p.useConfiguredOpenAICompatChat() { chatBody := p.buildOpenAICompatChatRequest(messages, tools, model, options) return p.postJSON(ctx, endpointFor(p.compatBase(), "/chat/completions"), chatBody) } @@ -309,7 +309,7 @@ func (p *HTTPProvider) callResponsesStream(ctx context.Context, messages []Messa if streamOpts, ok := mapOption(options, "responses_stream_options"); ok && len(streamOpts) > 0 { requestBody["stream_options"] = streamOpts } - if p.useOpenAICompatChatUpstream() { + if p.useOpenAICompatChatUpstream() || p.useConfiguredOpenAICompatChat() { chatBody := p.buildOpenAICompatChatRequest(messages, tools, model, options) chatBody["stream"] = true streamOptions := map[string]interface{}{"include_usage": true}