aicodeswitch 5.2.12 → 6.0.0

package/dist/server/proxy-server.js

@@ -54,6 +54,7 @@ const model_rewrite_transform_1 = require("./transformers/model-rewrite-transfor
 const chunk_collector_1 = require("./transformers/chunk-collector");
 const stream_timing_transform_1 = require("./transformers/stream-timing-transform");
 const rules_status_service_1 = require("./rules-status-service");
+const agent_map_1 = require("./agent-map");
 const index_1 = require("./conversions/index");
 const stream_converter_adapter_1 = require("./conversions/stream-converter-adapter");
 const types_1 = require("../types");
@@ -66,6 +67,13 @@ const coding_plan_1 = require("./coding-plan");
 const coding_plan_headers_1 = require("./coding-plan-headers");
 const auth_1 = require("./auth");
 const SUPPORTED_TARGETS = ['claude-code', 'codex'];
+/**
+ * Fallback（回退原始配置）路径的虚拟供应商归属。
+ * 该路径转发的请求不属于任何用户配置的供应商/服务，故用一组固定 ID + 名称
+ * 把它纳入服务性能统计（service-performance.json），便于在测速面板单独呈现。
+ */
+const FALLBACK_VENDOR_ID = 'fallback-vendor';
+const FALLBACK_VENDOR_NAME = '原始配置 / 直连';
 /** 默认模型列表 */
 const DEFAULT_MODELS = [
     { id: 'claude-sonnet-4-20250514', owned_by: 'anthropic' },
@@ -274,12 +282,16 @@ class ProxyServer {
      * 流式：依据 streamTiming 精确计算 TTFT 与生成阶段吞吐；非流式：端到端估算（estimated）。
      */
     emitPerformance(params) {
+        var _a;
         const tracker = this.performanceTracker;
         if (!tracker)
             return;
         const { statusCode, startTime, usage, streamTiming, service, vendorId, vendorName, model } = params;
         const isError = statusCode >= 400;
+        const inputTokens = usage === null || usage === void 0 ? void 0 : usage.inputTokens;
         const outputTokens = usage === null || usage === void 0 ? void 0 : usage.outputTokens;
+        const computedTotal = (inputTokens || 0) + (outputTokens || 0);
+        const totalTokens = (_a = usage === null || usage === void 0 ? void 0 : usage.totalTokens) !== null && _a !== void 0 ? _a : (computedTotal > 0 ? computedTotal : undefined);
         const responseMs = Date.now() - startTime;
         let ttftMs;
         let tokensPerSecond;
@@ -295,7 +307,11 @@ class ProxyServer {
         else if (outputTokens && responseMs > 0) {
             tokensPerSecond = outputTokens / (responseMs / 1000);
         }
-        tracker.recordPerformance(vendorId !== null && vendorId !== void 0 ? vendorId : service.vendorId, vendorName, service.id, service.name, model, { ttftMs, tokensPerSecond, outputTokens, timingAccuracy, isError });
+        // 解析最终归属：Fallback 临时服务没有真实 vendor，service.vendorId 兜底为虚拟供应商；
+        // 此时 vendorName 也为空，补上虚拟供应商名，确保不被 recordPerformance 的三元组校验丢弃。
+        const effectiveVendorId = vendorId !== null && vendorId !== void 0 ? vendorId : service.vendorId;
+        const effectiveVendorName = vendorName !== null && vendorName !== void 0 ? vendorName : (effectiveVendorId === FALLBACK_VENDOR_ID ? FALLBACK_VENDOR_NAME : undefined);
+        tracker.recordPerformance(effectiveVendorId, effectiveVendorName, service.id, service.name, model, { ttftMs, tokensPerSecond, outputTokens, inputTokens, totalTokens, timingAccuracy, isError });
     }
     /**
      * 从请求中提取 API Key（支持三种 Header，按优先级依次尝试）
@@ -749,6 +765,7 @@ class ProxyServer {
                     responseTime: Date.now() - requestStartAt,
                     // 添加最后失败的服务信息
                     ruleId: lastFailedRule === null || lastFailedRule === void 0 ? void 0 : lastFailedRule.id,
+                    routeId: route === null || route === void 0 ? void 0 : route.id,
                     targetServiceId: lastFailedService === null || lastFailedService === void 0 ? void 0 : lastFailedService.id,
                     targetServiceName: lastFailedService === null || lastFailedService === void 0 ? void 0 : lastFailedService.name,
                     targetModel: (lastFailedRule === null || lastFailedRule === void 0 ? void 0 : lastFailedRule.targetModel) || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
@@ -1258,6 +1275,7 @@ class ProxyServer {
                 const tempService = {
                     id: 'fallback-service',
                     name: 'Original Config',
+                    vendorId: FALLBACK_VENDOR_ID,
                     apiUrl: originalConfig.apiUrl,
                     apiKey: originalConfig.apiKey,
                     authType: originalConfig.authType,
@@ -3345,8 +3363,28 @@ class ProxyServer {
             });
         return { converter: adapter, extractUsage };
     }
+    /**
+     * 将 SSEEventCollector 归一化后的 usage（{input_tokens, output_tokens, total_tokens, cache_read_input_tokens}）
+     * 映射为 TokenUsage。collector 已遍历全部事件并合并 Anthropic message_start/message_delta + OpenAI/Gemini 字段，
+     * 因此这里无需再区分上游格式。
+     */
+    tokenUsageFromCollected(extracted) {
+        var _a;
+        if (!extracted)
+            return undefined;
+        const inputTokens = extracted.input_tokens || 0;
+        const outputTokens = extracted.output_tokens || 0;
+        const computedTotal = inputTokens + outputTokens;
+        const totalTokens = (_a = extracted.total_tokens) !== null && _a !== void 0 ? _a : (computedTotal > 0 ? computedTotal : undefined);
+        return {
+            inputTokens,
+            outputTokens,
+            totalTokens,
+            cacheReadInputTokens: extracted.cache_read_input_tokens || 0,
+        };
+    }
     extractTokenUsageFromResponse(responseData, sourceType) {
-        var _a, _b, _c, _d, _e, _f;
+        var _a, _b, _c, _d, _e, _f, _g, _h, _j;
         if (!responseData)
             return undefined;
         const format = (0, source_type_mapping_1.sourceTypeToFormat)(sourceType);
@@ -3362,14 +3400,15 @@ class ProxyServer {
             };
         }
         if (format === 'completions' || format === 'responses') {
-            const usage = responseData === null || responseData === void 0 ? void 0 : responseData.usage;
+            // 标准 Responses 非流式 usage 在顶层；个别上游/中转会嵌在 response.usage 下，做兜底
+            const usage = (_a = responseData === null || responseData === void 0 ? void 0 : responseData.usage) !== null && _a !== void 0 ? _a : (_b = responseData === null || responseData === void 0 ? void 0 : responseData.response) === null || _b === void 0 ? void 0 : _b.usage;
             if (!usage)
                 return undefined;
             return {
                 inputTokens: (usage === null || usage === void 0 ? void 0 : usage.input_tokens) || (usage === null || usage === void 0 ? void 0 : usage.prompt_tokens) || 0,
                 outputTokens: (usage === null || usage === void 0 ? void 0 : usage.output_tokens) || (usage === null || usage === void 0 ? void 0 : usage.completion_tokens) || 0,
                 totalTokens: (usage === null || usage === void 0 ? void 0 : usage.total_tokens) || 0,
-                cacheReadInputTokens: (usage === null || usage === void 0 ? void 0 : usage.cached_tokens) || (usage === null || usage === void 0 ? void 0 : usage.cache_read_input_tokens) || 0,
+                cacheReadInputTokens: (usage === null || usage === void 0 ? void 0 : usage.cached_tokens) || (usage === null || usage === void 0 ? void 0 : usage.cache_read_input_tokens) || ((_c = usage === null || usage === void 0 ? void 0 : usage.input_tokens_details) === null || _c === void 0 ? void 0 : _c.cached_tokens) || 0,
             };
         }
         if (format === 'claude') {
@@ -3377,7 +3416,7 @@ class ProxyServer {
                 return {
                     inputTokens: (responseData === null || responseData === void 0 ? void 0 : responseData.input_tokens) || 0,
                     outputTokens: (responseData === null || responseData === void 0 ? void 0 : responseData.output_tokens) || 0,
-                    totalTokens: ((_a = responseData === null || responseData === void 0 ? void 0 : responseData.input_tokens) !== null && _a !== void 0 ? _a : 0) + ((_b = responseData === null || responseData === void 0 ? void 0 : responseData.output_tokens) !== null && _b !== void 0 ? _b : 0) || undefined,
+                    totalTokens: ((_d = responseData === null || responseData === void 0 ? void 0 : responseData.input_tokens) !== null && _d !== void 0 ? _d : 0) + ((_e = responseData === null || responseData === void 0 ? void 0 : responseData.output_tokens) !== null && _e !== void 0 ? _e : 0) || undefined,
                     cacheReadInputTokens: (responseData === null || responseData === void 0 ? void 0 : responseData.cache_read_input_tokens) || 0,
                 };
             }
@@ -3387,7 +3426,7 @@ class ProxyServer {
             return {
                 inputTokens: (usage === null || usage === void 0 ? void 0 : usage.input_tokens) || 0,
                 outputTokens: (usage === null || usage === void 0 ? void 0 : usage.output_tokens) || 0,
-                totalTokens: ((_c = usage === null || usage === void 0 ? void 0 : usage.input_tokens) !== null && _c !== void 0 ? _c : 0) + ((_d = usage === null || usage === void 0 ? void 0 : usage.output_tokens) !== null && _d !== void 0 ? _d : 0) || undefined,
+                totalTokens: ((_f = usage === null || usage === void 0 ? void 0 : usage.input_tokens) !== null && _f !== void 0 ? _f : 0) + ((_g = usage === null || usage === void 0 ? void 0 : usage.output_tokens) !== null && _g !== void 0 ? _g : 0) || undefined,
                 cacheReadInputTokens: (usage === null || usage === void 0 ? void 0 : usage.cache_read_input_tokens) || 0,
             };
         }
@@ -3407,7 +3446,7 @@ class ProxyServer {
             return {
                 inputTokens: (usage === null || usage === void 0 ? void 0 : usage.input_tokens) || 0,
                 outputTokens: (usage === null || usage === void 0 ? void 0 : usage.output_tokens) || 0,
-                totalTokens: ((_e = usage === null || usage === void 0 ? void 0 : usage.input_tokens) !== null && _e !== void 0 ? _e : 0) + ((_f = usage === null || usage === void 0 ? void 0 : usage.output_tokens) !== null && _f !== void 0 ? _f : 0) || undefined,
+                totalTokens: ((_h = usage === null || usage === void 0 ? void 0 : usage.input_tokens) !== null && _h !== void 0 ? _h : 0) + ((_j = usage === null || usage === void 0 ? void 0 : usage.output_tokens) !== null && _j !== void 0 ? _j : 0) || undefined,
                 cacheReadInputTokens: (usage === null || usage === void 0 ? void 0 : usage.cache_read_input_tokens) || 0,
             };
         }
@@ -3415,7 +3454,7 @@ class ProxyServer {
     }
     proxyRequest(req, res, route, rule, service, options) {
         return __awaiter(this, void 0, void 0, function* () {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q, _r, _s;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q, _r, _s, _t, _u;
             res.locals.skipLog = true;
             const startTime = Date.now();
             const rawSourceType = service.sourceType || 'openai-chat';
@@ -3423,8 +3462,42 @@ class ProxyServer {
             const sourceType = (0, type_migration_1.normalizeSourceType)(rawSourceType);
             const targetType = this.inferToolFromRequest(req);
             const sessionId = this.defaultExtractSessionId(req, targetType) || '-';
+            // Agent Map：在途请求注册（active 状态判定依据）
+            const _accessKeyCtxAtStart = req._accessKeyCtx;
+            // 后台类请求（count_tokens / compact / background 规则）不重置「本轮通知标记」，避免主任务结束后的
+            // 续发请求重复触发结束通知
+            const _isBgRequest = rule.contentType === 'background'
+                || rule.contentType === 'compact'
+                || this.isCountTokensPath(req.path)
+                || this.isCountTokensPath(req.originalUrl);
+            // 思考类请求：复用路由同款识别（rule.contentType==='thinking' 或请求体带 thinking/reasoning 信号）。
+            // 思考期间用更宽的静默上限，避免思考静默被误判为「上游停滞」而弹通知。
+            const _isThinking = rule.contentType === 'thinking' || this.hasThinkingSignal(req.body);
+            res.locals._agentMapThinking = _isThinking;
+            agent_map_1.agentMapService.startRequest(sessionId, targetType, {
+                source: _accessKeyCtxAtStart ? 'access-key' : 'global',
+                keyId: (_a = _accessKeyCtxAtStart === null || _accessKeyCtxAtStart === void 0 ? void 0 : _accessKeyCtxAtStart.accessKey) === null || _a === void 0 ? void 0 : _a.id,
+                keyName: (_b = _accessKeyCtxAtStart === null || _accessKeyCtxAtStart === void 0 ? void 0 : _accessKeyCtxAtStart.accessKey) === null || _b === void 0 ? void 0 : _b.name,
+                background: _isBgRequest,
+                thinking: _isThinking,
+            });
+            // 安全网：无论走哪条分支（含编程套餐拒绝、配额拦截、异常等早退，导致 finalizeLog 未被调用），
+            // 都在响应关闭时确保 endRequest 配对执行一次，杜绝在途计数泄漏（「永远卡在进行中」）。
+            res.on('close', () => {
+                // 仅保证 endRequest 配对执行一次（防在途计数泄漏）。用独立标记 _agentMapEnded，
+                // 不再与 finalizeLog 的 onFinalized 共用标记 —— 否则 res 'close' 先于 finalizeLog 触发时
+                // 会把标记置位，导致 finalizeLog 跳过 onFinalized（requestCount / model / token 永不更新）。
+                if (res.locals._agentMapEnded)
+                    return;
+                res.locals._agentMapEnded = true;
+                agent_map_1.agentMapService.endRequest(sessionId, {
+                    isStream: !!res.locals._agentMapStream,
+                    thinking: !!res.locals._agentMapThinking,
+                });
+                agent_map_1.agentMapService.reevaluate(sessionId);
+            });
             const vendor = this.dbManager.getVendorByServiceId(service.id);
-            console.log(`\x1b[32m[Request Start]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, model=${rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model) || '-'}`);
+            console.log(`\x1b[32m[Request Start]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, model=${rule.targetModel || ((_c = req.body) === null || _c === void 0 ? void 0 : _c.model) || '-'}`);
             const failoverEnabled = (options === null || options === void 0 ? void 0 : options.failoverEnabled) === true;
             const forwardedToServiceName = options === null || options === void 0 ? void 0 : options.forwardedToServiceName;
             const useOriginalConfig = (options === null || options === void 0 ? void 0 : options.useOriginalConfig) === true;
@@ -3593,15 +3666,52 @@ class ProxyServer {
             // 标记规则正在使用
             rules_status_service_1.rulesStatusBroadcaster.markRuleInUse(route.id, rule.id);
             const finalizeLog = (statusCode, error) => __awaiter(this, void 0, void 0, function* () {
-                var _a, _b, _c, _d, _e, _f, _g, _h;
+                var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o;
                 if (logged)
                     return;
                 // 服务性能数据点采集（全局，与 AUTH 无关；独立于 enableLogging 开关）
                 this.emitPerformance({
                     statusCode, startTime, usage: usageForLog, streamTiming,
                     service, vendorId: vendor === null || vendor === void 0 ? void 0 : vendor.id, vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                    model: rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                    model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
                 });
+                // Agent Map：在途请求注销 + 活动/状态采集广播（独立于 enableLogging）
+                // 注意：endRequest 与 onFinalized 用各自独立的标记，二者解耦。
+                // 这样即便 res 'close' 安全网已先执行 endRequest（_agentMapEnded=true），onFinalized 仍会在此执行 ——
+                // 修复「requestCount / model / token 不更新」：旧实现二者共用 _agentMapRecorded，'close' 抢先置位会让此处整块跳过。
+                if (!res.locals._agentMapEnded) {
+                    res.locals._agentMapEnded = true;
+                    agent_map_1.agentMapService.endRequest(sessionId, {
+                        isStream: !!res.locals._agentMapStream,
+                        thinking: !!res.locals._agentMapThinking,
+                    });
+                }
+                if (!res.locals._agentMapFinalized) {
+                    res.locals._agentMapFinalized = true;
+                    const _akCtx = req._accessKeyCtx;
+                    const _tokensDelta = (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.totalTokens) ||
+                        (((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.inputTokens) || 0) + ((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.outputTokens) || 0));
+                    agent_map_1.agentMapService.onFinalized({
+                        sessionId,
+                        agent: targetType,
+                        source: _akCtx ? 'access-key' : 'global',
+                        keyId: (_b = _akCtx === null || _akCtx === void 0 ? void 0 : _akCtx.accessKey) === null || _b === void 0 ? void 0 : _b.id,
+                        keyName: (_c = _akCtx === null || _akCtx === void 0 ? void 0 : _akCtx.accessKey) === null || _c === void 0 ? void 0 : _c.name,
+                        title: this.defaultExtractSessionTitle(req, sessionId),
+                        timestamp: Date.now(),
+                        statusCode,
+                        // 采用真正转发给上游的模型（转换+覆盖后的 requestBody.model），而非 rule.targetModel 预测值，
+                        // 避免规则未配置 targetModel 时回退成编程工具提交的 req.body.model
+                        model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                        tokensDelta: _tokensDelta,
+                        // 输入/输出拆分（3D 连线两段用）；usageForLog 已规范化为驼峰字段
+                        inputTokensDelta: (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.inputTokens) || 0,
+                        outputTokensDelta: (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.outputTokens) || 0,
+                        body: req.body,
+                        downstreamResponseBody: downstreamResponseBodyForLog !== null && downstreamResponseBodyForLog !== void 0 ? downstreamResponseBodyForLog : responseBodyForLog,
+                        responseBody: responseBodyForLog,
+                    });
+                }
                 const isError = statusCode >= 400;
                 if (isError) {
                     console.log(`\x1b[31m[Request Error]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, status=${statusCode}, time=${Date.now() - startTime}ms${error ? `, error=${error}` : ''}`);
@@ -3610,7 +3720,7 @@ class ProxyServer {
                     console.log(`\x1b[33m[Request End]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, status=${statusCode}, time=${Date.now() - startTime}ms`);
                 }
                 // 检查是否启用日志记录（默认启用）
-                const enableLogging = ((_b = this.config) === null || _b === void 0 ? void 0 : _b.enableLogging) !== false; // 默认为 true
+                const enableLogging = ((_e = this.config) === null || _e === void 0 ? void 0 : _e.enableLogging) !== false; // 默认为 true
                 if (!enableLogging) {
                     return;
                 }
@@ -3634,13 +3744,14 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetType,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_c = req.body) === null || _c === void 0 ? void 0 : _c.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_f = req.body) === null || _f === void 0 ? void 0 : _f.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_d = req.body) === null || _d === void 0 ? void 0 : _d.model,
+                            requestModel: (_g = req.body) === null || _g === void 0 ? void 0 : _g.model,
                             tags: this.buildRelayTags(relayedForLog, useOriginalConfig),
                             responseHeaders: responseHeadersForLog,
                             responseBody: responseBodyForLog,
@@ -3674,7 +3785,7 @@ class ProxyServer {
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                                 serviceId: service.id,
                                 serviceName: service.name,
-                                model: rule.targetModel || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
+                                model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_h = req.body) === null || _h === void 0 ? void 0 : _h.model),
                                 totalTokens: sessionTokens,
                             }).catch(err => console.error('[KeySession] upsert error:', err));
                         }
@@ -3698,13 +3809,14 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetType,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_f = req.body) === null || _f === void 0 ? void 0 : _f.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_g = req.body) === null || _g === void 0 ? void 0 : _g.model,
+                            requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
                             tags: this.buildRelayTags(relayedForLog, useOriginalConfig),
                         });
                     }
@@ -3717,7 +3829,7 @@ class ProxyServer {
                 const vendors = this.dbManager.getVendors();
                 const vendorForLog = vendors.find(v => v.id === service.vendorId);
                 // 从请求体中提取模型信息
-                const requestModel = (_h = req.body) === null || _h === void 0 ? void 0 : _h.model;
+                const requestModel = (_l = req.body) === null || _l === void 0 ? void 0 : _l.model;
                 const tagsForLog = this.buildRelayTags(relayedForLog, useOriginalConfig);
                 if (extraTagsForLog.length > 0) {
                     tagsForLog.push(...extraTagsForLog);
@@ -3736,10 +3848,11 @@ class ProxyServer {
                     // 新增字段
                     contentType: rule.contentType,
                     ruleId: rule.id,
+                    routeId: route === null || route === void 0 ? void 0 : route.id,
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || requestModel,
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_m = req.body) === null || _m === void 0 ? void 0 : _m.model),
                     vendorId: service.vendorId,
                     vendorName: vendorForLog === null || vendorForLog === void 0 ? void 0 : vendorForLog.name,
                     requestModel,
@@ -3767,8 +3880,11 @@ class ProxyServer {
                         vendorName: vendorForLog === null || vendorForLog === void 0 ? void 0 : vendorForLog.name,
                         serviceId: service.id,
                         serviceName: service.name,
-                        model: requestModel || rule.targetModel,
+                        model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_o = req.body) === null || _o === void 0 ? void 0 : _o.model),
                         totalTokens,
+                        // 输入/输出拆分持久化（与 totalTokens 同口径），供 Agent Map 重启后恢复
+                        inputTokens: (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.inputTokens) || 0,
+                        outputTokens: (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.outputTokens) || 0,
                         highIqMode: rule.contentType === 'high-iq' ? true : existingSession === null || existingSession === void 0 ? void 0 : existingSession.highIqMode,
                         highIqRuleId: rule.contentType === 'high-iq' ? rule.id : existingSession === null || existingSession === void 0 ? void 0 : existingSession.highIqRuleId,
                         highIqEnabledAt: rule.contentType === 'high-iq'
@@ -3868,7 +3984,7 @@ class ProxyServer {
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
                     vendorId: service.vendorId,
                     vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                     requestModel: (_b = req.body) === null || _b === void 0 ? void 0 : _b.model,
@@ -3934,7 +4050,7 @@ class ProxyServer {
                 // responses→responses 直连非 OpenAI 官方端点时，需降级兼容（剥离 custom/namespace 等私有工具与非标准字段）
                 const sanitizeBody = clientFormat === 'responses' && (0, source_type_mapping_1.sourceTypeToFormat)(sourceType) === 'responses' && !(0, index_1.isOfficialOpenAiApi)(effectiveApiUrl || '');
                 const transformedRequestBody = this.transformRequestToUpstream(targetType, sourceType, payloadForTransform, rule.targetModel, providerConfig, serverToolConfig, sanitizeBody);
-                requestBody = (_b = transformedRequestBody !== null && transformedRequestBody !== void 0 ? transformedRequestBody : this.cloneRequestBody(originalToolRequestBody)) !== null && _b !== void 0 ? _b : {};
+                requestBody = (_d = transformedRequestBody !== null && transformedRequestBody !== void 0 ? transformedRequestBody : this.cloneRequestBody(originalToolRequestBody)) !== null && _d !== void 0 ? _d : {};
                 // 对最终即将发送到上游的 Claude compact 请求再做一次兜底清理，
                 // 避免中间转换/覆盖步骤重新引入未配对的 tool_use。
                 if (rule.contentType === 'compact' && targetType === 'claude-code' && Array.isArray(requestBody === null || requestBody === void 0 ? void 0 : requestBody.messages)) {
@@ -4040,7 +4156,7 @@ class ProxyServer {
                     let errorResponseData = response.data;
                     if (streamRequested && response.data && typeof response.data.on === 'function') {
                         const raw = yield this.readStreamBody(response.data);
-                        errorResponseData = (_c = this.safeJsonParse(raw)) !== null && _c !== void 0 ? _c : raw;
+                        errorResponseData = (_e = this.safeJsonParse(raw)) !== null && _e !== void 0 ? _e : raw;
                     }
                     responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
                     yield handleUpstreamHttpError(response.status, errorResponseData, responseHeaders, contentType);
@@ -4078,10 +4194,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_f = req.body) === null || _f === void 0 ? void 0 : _f.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_e = req.body) === null || _e === void 0 ? void 0 : _e.model,
+                                requestModel: (_g = req.body) === null || _g === void 0 ? void 0 : _g.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4112,18 +4228,28 @@ class ProxyServer {
                     const serializer = new streaming_1.SSESerializerTransform();
                     const downstreamChunkCollector = new chunk_collector_1.ChunkCollectorTransform(() => {
                         rules_status_service_1.rulesStatusBroadcaster.refreshRuleInUse(route.id, rule.id);
+                        // Agent Map 心跳：每个流经的下游 chunk 都视为「会话仍活跃」，刷新活动时钟（节流由 ChunkCollector 内部 5s 上限控制）
+                        // 同时把当前流式请求的实时累计 usage（input/output）一并推送，使前端节点在流式过程中随 token 增长实时上移
+                        const u = eventCollector.extractUsage();
+                        agent_map_1.agentMapService.heartbeat(sessionId, u ? {
+                            inputTokens: u.input_tokens || 0,
+                            outputTokens: u.output_tokens || 0,
+                        } : undefined);
                     });
                     // 服务性能打点：记录首/末 SSE 事件时间，用于 TTFT 与生成阶段吞吐
                     streamTiming = new stream_timing_transform_1.StreamTimingTransform(startTime);
+                    // Agent Map：SSE 流已建立 —— 记一个在途流，状态机据此区分「同步/流式」活跃判定
+                    res.locals._agentMapStream = true;
+                    agent_map_1.agentMapService.markStreaming(sessionId);
                     const compactResponseSanitizer = rule.contentType === 'compact' && targetType === 'claude-code'
                         ? new ClaudeCompactResponseSanitizer()
                         : null;
                     // 流式 model 回写：将上游返回的 model 改写为客户端请求时的原始模型名
-                    const originalModel = (_f = req.body) === null || _f === void 0 ? void 0 : _f.model;
+                    const originalModel = (_h = req.body) === null || _h === void 0 ? void 0 : _h.model;
                     const modelRewriter = originalModel ? new model_rewrite_transform_1.ModelRewriteTransform(originalModel) : null;
                     responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
                     // 使用 transformSSEToTool 方法选择转换器
-                    const { converter, extractUsage } = this.transformSSEToTool(targetType, sourceType);
+                    const { converter } = this.transformSSEToTool(targetType, sourceType);
                     this.copyResponseHeaders(responseHeaders, res);
                     // 收集日志：responseBody/streamChunks 记录上游原始响应；downstreamResponseBody 记录实际下发内容
                     const finalizeChunks = () => {
@@ -4140,13 +4266,7 @@ class ProxyServer {
                                 extractedUsage = converterUsage || extractedUsage;
                             }
                         }
-                        // 如果有自定义的 extractUsage 函数，使用它
-                        if (extractUsage && extractedUsage) {
-                            usageForLog = extractUsage(extractedUsage);
-                        }
-                        else if (extractedUsage) {
-                            usageForLog = this.extractTokenUsageFromResponse(extractedUsage, sourceType);
-                        }
+                        usageForLog = this.tokenUsageFromCollected(extractedUsage);
                     };
                     // 监听 res 的错误事件
                     res.on('error', (err) => {
@@ -4219,10 +4339,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_g = req.body) === null || _g === void 0 ? void 0 : _g.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_h = req.body) === null || _h === void 0 ? void 0 : _h.model,
+                                requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4266,10 +4386,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_l = req.body) === null || _l === void 0 ? void 0 : _l.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
+                                requestModel: (_m = req.body) === null || _m === void 0 ? void 0 : _m.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4288,7 +4408,7 @@ class ProxyServer {
                 let responseData = response.data;
                 if (streamRequested && response.data && typeof response.data.on === 'function' && !isEventStream) {
                     const raw = yield this.readStreamBody(response.data);
-                    responseData = (_l = this.safeJsonParse(raw)) !== null && _l !== void 0 ? _l : raw;
+                    responseData = (_o = this.safeJsonParse(raw)) !== null && _o !== void 0 ? _o : raw;
                 }
                 // 收集响应头
                 responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
@@ -4310,7 +4430,7 @@ class ProxyServer {
                 usageForLog = this.extractTokenUsageFromResponse(responseData, sourceType);
                 console.log('[Proxy] Non-stream response: extracted usageForLog:', usageForLog);
                 // 回写 model 字段：将上游返回的 model 改写为客户端请求时的原始模型名
-                const originalModel = (_m = req.body) === null || _m === void 0 ? void 0 : _m.model;
+                const originalModel = (_p = req.body) === null || _p === void 0 ? void 0 : _p.model;
                 (0, model_rewrite_transform_1.rewriteResponseModel)(normalizedConverted, originalModel);
                 (0, model_rewrite_transform_1.rewriteResponseModel)(responseData, originalModel);
                 this.copyResponseHeaders(responseHeaders, res);
@@ -4375,10 +4495,10 @@ class ProxyServer {
                         targetType,
                         targetServiceId: service.id,
                         targetServiceName: service.name,
-                        targetModel: rule.targetModel || ((_o = req.body) === null || _o === void 0 ? void 0 : _o.model),
+                        targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_q = req.body) === null || _q === void 0 ? void 0 : _q.model),
                         vendorId: service.vendorId,
                         vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                        requestModel: (_p = req.body) === null || _p === void 0 ? void 0 : _p.model,
+                        requestModel: (_r = req.body) === null || _r === void 0 ? void 0 : _r.model,
                         upstreamRequest: upstreamRequestForLog,
                         responseTime: Date.now() - startTime,
                     });
@@ -4393,7 +4513,7 @@ class ProxyServer {
                 console.error('Proxy error:', error);
                 // 检测是否是 timeout 错误
                 const isTimeout = error.code === 'ECONNABORTED' ||
-                    ((_q = error.message) === null || _q === void 0 ? void 0 : _q.toLowerCase().includes('timeout')) ||
+                    ((_s = error.message) === null || _s === void 0 ? void 0 : _s.toLowerCase().includes('timeout')) ||
                     (error.errno && error.errno === 'ETIMEDOUT');
                 const statusCode = isTimeout ? 504 : this.getErrorStatusCode(error, 500);
                 const baseErrorMessage = isTimeout
@@ -4419,10 +4539,10 @@ class ProxyServer {
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || ((_r = req.body) === null || _r === void 0 ? void 0 : _r.model),
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_t = req.body) === null || _t === void 0 ? void 0 : _t.model),
                     vendorId: service.vendorId,
                     vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                    requestModel: (_s = req.body) === null || _s === void 0 ? void 0 : _s.model,
+                    requestModel: (_u = req.body) === null || _u === void 0 ? void 0 : _u.model,
                     upstreamRequest: upstreamRequestForLog,
                     responseHeaders: responseHeadersForLog,
                     responseTime: Date.now() - startTime,
@@ -4693,7 +4813,7 @@ class ProxyServer {
                 this.emitPerformance({
                     statusCode, startTime, usage: usageForLog, streamTiming,
                     service, vendorId: vendor === null || vendor === void 0 ? void 0 : vendor.id, vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                    model: rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                    model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
                 });
                 // AccessKey 独立日志处理
                 const accessKeyCtx = req._accessKeyCtx;
@@ -4711,9 +4831,10 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_b = req.body) === null || _b === void 0 ? void 0 : _b.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_b = req.body) === null || _b === void 0 ? void 0 : _b.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                             requestModel: (_c = req.body) === null || _c === void 0 ? void 0 : _c.model,
@@ -4742,7 +4863,7 @@ class ProxyServer {
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                                 serviceId: service.id,
                                 serviceName: service.name,
-                                model: rule.targetModel || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                                model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
                                 totalTokens: sessionTokens,
                             }).catch(err => console.error('[KeySession] upsert error:', err));
                         }
@@ -4764,9 +4885,10 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                             requestModel: (_f = req.body) === null || _f === void 0 ? void 0 : _f.model,
@@ -4945,7 +5067,7 @@ class ProxyServer {
                     // 流式 model 回写：将上游返回的 model 改写为客户端请求时的原始模型名
                     const originalModel = (_d = req.body) === null || _d === void 0 ? void 0 : _d.model;
                     const modelRewriter = originalModel ? new model_rewrite_transform_1.ModelRewriteTransform(originalModel) : null;
-                    const { converter, extractUsage } = this.transformSSEByFormat(clientFormat, sourceType);
+                    const { converter } = this.transformSSEByFormat(clientFormat, sourceType);
                     this.copyResponseHeaders(responseHeaders, res);
                     res.status(response.status);
                     const finalizeStreamChunks = () => {
@@ -4958,12 +5080,7 @@ class ProxyServer {
                             if (converterUsage)
                                 extractedUsage = converterUsage;
                         }
-                        if (extractUsage && extractedUsage) {
-                            usageForLog = extractUsage(extractedUsage);
-                        }
-                        else if (extractedUsage) {
-                            usageForLog = this.extractTokenUsageFromResponse(extractedUsage, sourceType);
-                        }
+                        usageForLog = this.tokenUsageFromCollected(extractedUsage);
                     };
                     try {
                         yield new Promise((resolve, reject) => {