feat(provider): support chat completions for openai providers

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：

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:

cmd/cmd_gateway_test.go

@@ -4,9 +4,12 @@ import (
 	"context"
 	"os"
 	"path/filepath"
+	"reflect"
 	"sync/atomic"
 	"testing"
 	"time"
+
+	"github.com/YspCoder/clawgo/pkg/config"
 )
 
 func TestConfigFileFingerprintSameContentIgnoresTouch(t *testing.T) {
@@ -115,3 +118,43 @@ func TestGatewayConfigWatcherTouchDoesNotReload(t *testing.T) {
 		t.Fatalf("expected touch-only update to skip reload, got %d", got)
 	}
 }
+
+func TestNormalizeHotReloadChannelsConfigIgnoresWeixinRuntimeState(t *testing.T) {
+	t.Parallel()
+
+	base := config.ChannelsConfig{
+		Weixin: config.WeixinConfig{
+			Enabled:      true,
+			BaseURL:      "https://ilinkai.weixin.qq.com",
+			DefaultBotID: "bot-a",
+			Accounts: []config.WeixinAccountConfig{
+				{
+					BotID:         "bot-a",
+					BotToken:      "token-a",
+					IlinkUserID:   "u-1",
+					ContextToken:  "ctx-a",
+					GetUpdatesBuf: "buf-a",
+				},
+			},
+			ContextToken:  "root-ctx",
+			GetUpdatesBuf: "root-buf",
+		},
+	}
+	next := base
+	next.Weixin.ContextToken = "root-ctx-next"
+	next.Weixin.GetUpdatesBuf = "root-buf-next"
+	next.Weixin.Accounts[0].ContextToken = "ctx-b"
+	next.Weixin.Accounts[0].GetUpdatesBuf = "buf-b"
+
+	left := normalizeHotReloadChannelsConfig(base)
+	right := normalizeHotReloadChannelsConfig(next)
+	if !reflect.DeepEqual(left, right) {
+		t.Fatalf("expected weixin runtime state changes to be ignored during hot reload comparison")
+	}
+
+	next.Weixin.BaseURL = "https://redirect.example"
+	right = normalizeHotReloadChannelsConfig(next)
+	if reflect.DeepEqual(left, right) {
+		t.Fatalf("expected durable weixin config changes to remain visible to hot reload comparison")
+	}
+}

cmd/gateway_reload.go

@@ -67,10 +67,12 @@ func (r *gatewayReloader) trigger(source string, forceRuntimeReload bool) error
 			r.state.cfg.Gateway.Host, r.state.cfg.Gateway.Port, newCfg.Gateway.Host, newCfg.Gateway.Port)
 	}
 
+	currentChannels := normalizeHotReloadChannelsConfig(r.state.cfg.Channels)
+	nextChannels := normalizeHotReloadChannelsConfig(newCfg.Channels)
 	runtimeSame := reflect.DeepEqual(r.state.cfg.Agents, newCfg.Agents) &&
 		reflect.DeepEqual(r.state.cfg.Models, newCfg.Models) &&
 		reflect.DeepEqual(r.state.cfg.Tools, newCfg.Tools) &&
-		reflect.DeepEqual(r.state.cfg.Channels, newCfg.Channels)
+		reflect.DeepEqual(currentChannels, nextChannels)
 
 	if runtimeSame && !forceRuntimeReload {
 		configureLogging(newCfg)
@@ -146,6 +148,16 @@ func (r *gatewayReloader) bindWeixinChannel() {
 	}
 }
 
+func normalizeHotReloadChannelsConfig(cfg config.ChannelsConfig) config.ChannelsConfig {
+	cfg.Weixin.ContextToken = ""
+	cfg.Weixin.GetUpdatesBuf = ""
+	for i := range cfg.Weixin.Accounts {
+		cfg.Weixin.Accounts[i].ContextToken = ""
+		cfg.Weixin.Accounts[i].GetUpdatesBuf = ""
+	}
+	return cfg
+}
+
 type configFileFingerprint struct {
 	Size        int64
 	ModUnixNano int64

cmd/main.go

@@ -15,7 +15,7 @@ import (
 	"github.com/YspCoder/clawgo/pkg/logger"
 )
 
-var version = "1.2.2"
+var version = "1.2.3"
 var buildTime = "unknown"
 
 const logo = ">"

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": [],

pkg/channels/dedupe_regression_test.go

@@ -33,6 +33,21 @@ func (r *recordingChannel) count() int {
 	return len(r.sent)
 }
 
+type canceledChannel struct {
+	called chan struct{}
+}
+
+func (c *canceledChannel) Name() string                          { return "test" }
+func (c *canceledChannel) Start(ctx context.Context) error       { return nil }
+func (c *canceledChannel) Stop(ctx context.Context) error        { return nil }
+func (c *canceledChannel) IsRunning() bool                       { return true }
+func (c *canceledChannel) IsAllowed(senderID string) bool        { return true }
+func (c *canceledChannel) HealthCheck(ctx context.Context) error { return nil }
+func (c *canceledChannel) Send(ctx context.Context, msg bus.OutboundMessage) error {
+	close(c.called)
+	return context.Canceled
+}
+
 func TestDispatchOutbound_DeduplicatesRepeatedSend(t *testing.T) {
 	mb := bus.NewMessageBus()
 	mgr, err := NewManager(&config.Config{}, mb)
@@ -58,6 +73,28 @@ func TestDispatchOutbound_DeduplicatesRepeatedSend(t *testing.T) {
 	}
 }
 
+func TestDispatchOutbound_TreatsCanceledSendAsLifecycleExit(t *testing.T) {
+	mb := bus.NewMessageBus()
+	mgr, err := NewManager(&config.Config{}, mb)
+	if err != nil {
+		t.Fatalf("new manager: %v", err)
+	}
+	cc := &canceledChannel{called: make(chan struct{})}
+	mgr.channels["test"] = cc
+	mgr.refreshSnapshot()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go mgr.dispatchOutbound(ctx)
+
+	mb.PublishOutbound(bus.OutboundMessage{Channel: "test", ChatID: "c1", Content: "hello", Action: "send"})
+	select {
+	case <-cc.called:
+	case <-time.After(time.Second):
+		t.Fatalf("expected canceled send to be dispatched")
+	}
+}
+
 func TestBaseChannel_HandleMessage_ContentHashFallbackDedupe(t *testing.T) {
 	mb := bus.NewMessageBus()
 	bc := NewBaseChannel("test", nil, mb, nil)

pkg/channels/manager.go

@@ -9,6 +9,7 @@ package channels
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"hash/fnv"
 	"strings"
@@ -339,6 +340,13 @@ func (m *Manager) dispatchOutbound(ctx context.Context) {
 			go func(c Channel, outbound bus.OutboundMessage) {
 				defer func() { <-m.dispatchSem }()
 				if err := c.Send(ctx, outbound); err != nil {
+					if errors.Is(err, context.Canceled) {
+						logger.InfoCF("channels", logger.C0042, map[string]interface{}{
+							logger.FieldChannel: outbound.Channel,
+							"reason":            "context canceled",
+						})
+						return
+					}
 					logger.ErrorCF("channels", logger.C0042, map[string]interface{}{
 						logger.FieldChannel: outbound.Channel,
 						logger.FieldError:   err.Error(),

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"`

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)
+	}
 }

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":

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)
+	}
+}

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)

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:

pkg/providers/openai_compat_provider.go

@@ -202,14 +202,5 @@ func applyAttemptFailure(base *HTTPProvider, attempt authAttempt, reason oauthFa
 }
 
 func estimateOpenAICompatTokenCount(body map[string]interface{}) (int, error) {
-	data, err := json.Marshal(body)
-	if err != nil {
-		return 0, fmt.Errorf("failed to encode request for token count: %w", err)
-	}
-	const charsPerToken = 4
-	count := (len(data) + charsPerToken - 1) / charsPerToken
-	if count < 1 {
-		count = 1
-	}
-	return count, nil
+	return EstimateOpenAICompatRequestTokens(body)
 }

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)
+	}
+}

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 {

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}

pkg/providers/token_estimator.go

@@ -0,0 +1,316 @@
+package providers
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"strings"
+	"unicode"
+	"unicode/utf8"
+)
+
+const (
+	estimateMessageOverheadTokens = 4
+	estimateNameOverheadTokens    = 1
+	estimateToolCallOverhead      = 8
+	estimateToolDefOverhead       = 10
+	estimateImageLowTokens        = 85
+	estimateImageHighTokens       = 255
+	estimateFileTokens            = 120
+)
+
+// EstimateOpenAICompatRequestTokens estimates prompt tokens for an OpenAI-compatible
+// chat request without calling an upstream tokenizer. It intentionally errs a bit
+// high for structured fields so compaction triggers before providers reject a prompt.
+func EstimateOpenAICompatRequestTokens(body map[string]interface{}) (int, error) {
+	if body == nil {
+		return 1, nil
+	}
+	count := 0
+	if model := strings.TrimSpace(asEstimateString(body["model"])); model != "" {
+		count += estimateTextTokens(model)
+	}
+	count += estimateOpenAICompatMessages(body["messages"])
+	count += estimateOpenAICompatTools(body["tools"])
+	count += estimateReasoningTokens(body)
+	count += estimateGenericOptions(body)
+	if count < 1 {
+		count = 1
+	}
+	return count, nil
+}
+
+// EstimatePromptTokens estimates tokens directly from provider-native message and
+// tool structures. Providers with native count APIs should still prefer those.
+func EstimatePromptTokens(messages []Message, tools []ToolDefinition, model string, options map[string]interface{}) int {
+	body := map[string]interface{}{
+		"model":    model,
+		"messages": openAICompatMessages(messages),
+	}
+	if len(tools) > 0 {
+		body["tools"] = openAICompatTools(tools)
+	}
+	for key, value := range options {
+		body[key] = value
+	}
+	count, err := EstimateOpenAICompatRequestTokens(body)
+	if err != nil {
+		return 1
+	}
+	return count
+}
+
+func estimateOpenAICompatMessages(raw interface{}) int {
+	messages, ok := raw.([]map[string]interface{})
+	if !ok {
+		if arr, ok := raw.([]interface{}); ok {
+			total := 0
+			for _, item := range arr {
+				if msg, ok := item.(map[string]interface{}); ok {
+					total += estimateOpenAICompatMessage(msg)
+				}
+			}
+			return total
+		}
+		return estimateJSONTokens(raw)
+	}
+	total := 0
+	for _, msg := range messages {
+		total += estimateOpenAICompatMessage(msg)
+	}
+	return total
+}
+
+func estimateOpenAICompatMessage(msg map[string]interface{}) int {
+	if msg == nil {
+		return 0
+	}
+	total := estimateMessageOverheadTokens
+	total += estimateTextTokens(asEstimateString(msg["role"]))
+	if name := strings.TrimSpace(asEstimateString(msg["name"])); name != "" {
+		total += estimateNameOverheadTokens + estimateTextTokens(name)
+	}
+	if toolCallID := strings.TrimSpace(asEstimateString(msg["tool_call_id"])); toolCallID != "" {
+		total += estimateTextTokens(toolCallID)
+	}
+	total += estimateContentTokens(msg["content"])
+	total += estimateToolCalls(msg["tool_calls"])
+	return total
+}
+
+func estimateContentTokens(content interface{}) int {
+	switch v := content.(type) {
+	case nil:
+		return 0
+	case string:
+		return estimateTextTokens(v)
+	case []map[string]interface{}:
+		total := 0
+		for _, part := range v {
+			total += estimateContentPartTokens(part)
+		}
+		return total
+	case []interface{}:
+		total := 0
+		for _, raw := range v {
+			if part, ok := raw.(map[string]interface{}); ok {
+				total += estimateContentPartTokens(part)
+				continue
+			}
+			total += estimateJSONTokens(raw)
+		}
+		return total
+	default:
+		return estimateJSONTokens(v)
+	}
+}
+
+func estimateContentPartTokens(part map[string]interface{}) int {
+	if part == nil {
+		return 0
+	}
+	typ := strings.ToLower(strings.TrimSpace(asEstimateString(part["type"])))
+	switch typ {
+	case "text", "input_text":
+		return estimateTextTokens(asEstimateString(part["text"]))
+	case "image_url", "input_image":
+		detail := strings.ToLower(strings.TrimSpace(asEstimateString(part["detail"])))
+		if detail == "" {
+			if image, ok := part["image_url"].(map[string]interface{}); ok {
+				detail = strings.ToLower(strings.TrimSpace(asEstimateString(image["detail"])))
+			}
+		}
+		if detail == "low" {
+			return estimateImageLowTokens
+		}
+		return estimateImageHighTokens
+	case "input_file", "file":
+		return estimateFileTokens + estimateJSONTokens(part)
+	default:
+		if text := strings.TrimSpace(asEstimateString(part["text"])); text != "" {
+			return estimateTextTokens(text)
+		}
+		return estimateJSONTokens(part)
+	}
+}
+
+func estimateToolCalls(raw interface{}) int {
+	calls, ok := raw.([]map[string]interface{})
+	if !ok {
+		if arr, ok := raw.([]interface{}); ok {
+			total := 0
+			for _, item := range arr {
+				if call, ok := item.(map[string]interface{}); ok {
+					total += estimateToolCall(call)
+				}
+			}
+			return total
+		}
+		return 0
+	}
+	total := 0
+	for _, call := range calls {
+		total += estimateToolCall(call)
+	}
+	return total
+}
+
+func estimateToolCall(call map[string]interface{}) int {
+	if call == nil {
+		return 0
+	}
+	total := estimateToolCallOverhead
+	total += estimateTextTokens(asEstimateString(call["id"]))
+	total += estimateTextTokens(asEstimateString(call["type"]))
+	if fn, ok := call["function"].(map[string]interface{}); ok {
+		total += estimateTextTokens(asEstimateString(fn["name"]))
+		total += estimateTextTokens(asEstimateString(fn["arguments"]))
+	}
+	total += estimateTextTokens(asEstimateString(call["name"]))
+	total += estimateJSONTokens(call["arguments"])
+	return total
+}
+
+func estimateOpenAICompatTools(raw interface{}) int {
+	tools, ok := raw.([]map[string]interface{})
+	if !ok {
+		if arr, ok := raw.([]interface{}); ok {
+			total := 0
+			for _, item := range arr {
+				if tool, ok := item.(map[string]interface{}); ok {
+					total += estimateToolDefinition(tool)
+				}
+			}
+			return total
+		}
+		return 0
+	}
+	total := 0
+	for _, tool := range tools {
+		total += estimateToolDefinition(tool)
+	}
+	return total
+}
+
+func estimateToolDefinition(tool map[string]interface{}) int {
+	if tool == nil {
+		return 0
+	}
+	total := estimateToolDefOverhead + estimateTextTokens(asEstimateString(tool["type"]))
+	if fn, ok := tool["function"].(map[string]interface{}); ok {
+		total += estimateTextTokens(asEstimateString(fn["name"]))
+		total += estimateTextTokens(asEstimateString(fn["description"]))
+		total += estimateJSONTokens(fn["parameters"])
+		if strict, ok := fn["strict"]; ok {
+			total += estimateJSONTokens(strict)
+		}
+		return total
+	}
+	total += estimateTextTokens(asEstimateString(tool["name"]))
+	total += estimateTextTokens(asEstimateString(tool["description"]))
+	total += estimateJSONTokens(tool["parameters"])
+	return total
+}
+
+func estimateReasoningTokens(body map[string]interface{}) int {
+	total := 0
+	if effort := strings.ToLower(strings.TrimSpace(asEstimateString(body["reasoning_effort"]))); effort != "" {
+		total += estimateTextTokens(effort)
+		switch effort {
+		case "minimal", "low":
+			total += 32
+		case "medium", "auto":
+			total += 96
+		case "high":
+			total += 192
+		}
+	}
+	for _, key := range []string{"reasoning", "chat_template_kwargs"} {
+		if value, ok := body[key]; ok {
+			total += estimateJSONTokens(value)
+		}
+	}
+	return total
+}
+
+func estimateGenericOptions(body map[string]interface{}) int {
+	total := 0
+	for _, key := range []string{"tool_choice", "parallel_tool_calls", "response_format"} {
+		if value, ok := body[key]; ok {
+			total += estimateJSONTokens(value)
+		}
+	}
+	return total
+}
+
+func estimateJSONTokens(value interface{}) int {
+	if value == nil {
+		return 0
+	}
+	data, err := json.Marshal(value)
+	if err != nil {
+		return estimateTextTokens(fmt.Sprintf("%v", value))
+	}
+	return estimateTextTokens(string(data))
+}
+
+func estimateTextTokens(text string) int {
+	text = strings.TrimSpace(text)
+	if text == "" {
+		return 0
+	}
+	runes := utf8.RuneCountInString(text)
+	ascii := 0
+	han := 0
+	other := 0
+	for _, r := range text {
+		switch {
+		case r <= unicode.MaxASCII:
+			ascii++
+		case unicode.Is(unicode.Han, r):
+			han++
+		default:
+			other++
+		}
+	}
+	asciiTokens := int(math.Ceil(float64(ascii) / 4.0))
+	otherTokens := int(math.Ceil(float64(other) / 2.0))
+	total := asciiTokens + han + otherTokens
+	if total < 1 && runes > 0 {
+		return 1
+	}
+	return total
+}
+
+func asEstimateString(value interface{}) string {
+	switch v := value.(type) {
+	case nil:
+		return ""
+	case string:
+		return v
+	case fmt.Stringer:
+		return v.String()
+	default:
+		return fmt.Sprintf("%v", v)
+	}
+}

pkg/providers/token_estimator_test.go

@@ -0,0 +1,72 @@
+package providers
+
+import "testing"
+
+func TestEstimatePromptTokensCountsMessagesToolsAndToolCalls(t *testing.T) {
+	tools := []ToolDefinition{{
+		Type: "function",
+		Function: ToolFunctionDefinition{
+			Name:        "lookup_weather",
+			Description: "Look up weather by city.",
+			Parameters: map[string]interface{}{
+				"type": "object",
+				"properties": map[string]interface{}{
+					"city": map[string]interface{}{"type": "string"},
+				},
+				"required": []string{"city"},
+			},
+		},
+	}}
+	messages := []Message{
+		{Role: "system", Content: "You are concise."},
+		{Role: "user", Content: "北京天气怎么样"},
+		{
+			Role: "assistant",
+			ToolCalls: []ToolCall{{
+				ID:   "call_1",
+				Type: "function",
+				Function: &FunctionCall{
+					Name:      "lookup_weather",
+					Arguments: `{"city":"北京"}`,
+				},
+			}},
+		},
+	}
+
+	withoutTools := EstimatePromptTokens(messages, nil, "qwen-max", nil)
+	withTools := EstimatePromptTokens(messages, tools, "qwen-max", map[string]interface{}{"reasoning_effort": "medium"})
+
+	if withoutTools <= 0 {
+		t.Fatalf("withoutTools = %d, want positive estimate", withoutTools)
+	}
+	if withTools <= withoutTools {
+		t.Fatalf("withTools = %d, want > withoutTools %d", withTools, withoutTools)
+	}
+}
+
+func TestEstimateOpenAICompatRequestTokensCountsMultimodalParts(t *testing.T) {
+	base := NewHTTPProvider("openai", "token", "https://example.com/v1", "gpt-5", false, "api_key", 5, nil)
+	textOnly := base.buildOpenAICompatChatRequest([]Message{{
+		Role:    "user",
+		Content: "look",
+	}}, nil, "gpt-5", nil)
+	withImage := base.buildOpenAICompatChatRequest([]Message{{
+		Role: "user",
+		ContentParts: []MessageContentPart{
+			{Type: "input_text", Text: "look"},
+			{Type: "input_image", ImageURL: "https://example.com/cat.png", Detail: "high"},
+		},
+	}}, nil, "gpt-5", nil)
+
+	textCount, err := EstimateOpenAICompatRequestTokens(textOnly)
+	if err != nil {
+		t.Fatalf("text estimate error: %v", err)
+	}
+	imageCount, err := EstimateOpenAICompatRequestTokens(withImage)
+	if err != nil {
+		t.Fatalf("image estimate error: %v", err)
+	}
+	if imageCount < textCount+estimateImageHighTokens {
+		t.Fatalf("imageCount = %d, textCount = %d, want image overhead", imageCount, textCount)
+	}
+}