aicodeswitch 5.2.10 → 5.2.12

package/bin/utils/managed-fields.js

@@ -45,6 +45,8 @@ const CODEX_CONFIG_MANAGED_FIELDS = [
   'enableRouteSelection',
   'model_providers.aicodeswitch',
   'mcp_servers',
+  'features',
+  'memories',
 ];
 
 /**

package/dist/server/config-managed-fields.js

@@ -41,6 +41,8 @@ exports.CODEX_CONFIG_MANAGED_FIELDS = [
     { path: ['enableRouteSelection'] },
     { path: ['model_providers', 'aicodeswitch'], isSection: true },
     { path: ['mcp_servers'], isSection: true, optional: true },
+    { path: ['features'], isSection: true, optional: true },
+    { path: ['memories'], isSection: true, optional: true },
 ];
 /**
  * Codex auth.json 管理字段定义

package/dist/server/conversions/pairs/responses-claude/response.js

@@ -9,6 +9,7 @@ exports.claudeToResponsesResponse = claudeToResponsesResponse;
 const id_js_1 = require("../../utils/id.js");
 const stop_reasons_js_1 = require("../../utils/stop-reasons.js");
 const mapper_js_1 = require("../../thinking/mapper.js");
+const usage_js_1 = require("../../utils/usage.js");
 /**
  * Convert a Claude Messages response to an OpenAI Responses API response.
  */
@@ -45,26 +46,9 @@ function claudeToResponsesResponse(response) {
         }
     }
     const { status, incomplete_details } = (0, stop_reasons_js_1.claudeToResponsesStatus)(response.stop_reason);
-    const usage = claudeToResponsesUsage(response.usage);
+    // 上游无 usage 时省略 usage 字段（不伪造 0）
+    const usage = (0, usage_js_1.toResponsesUsage)(response.usage);
     const responseId = response.id || (0, id_js_1.generateResponseId)();
-    return Object.assign(Object.assign({ id: responseId, object: 'response', status,
-        output, model: response.model || '', created_at: Math.floor(Date.now() / 1000), usage }, (incomplete_details ? { incomplete_details } : {})), { metadata: {} });
-}
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-/**
- * Map Claude usage to Responses API usage.
- */
-function claudeToResponsesUsage(usage) {
-    var _a, _b;
-    if (!usage)
-        return { input_tokens: 0, output_tokens: 0, total_tokens: 0 };
-    const input = (_a = usage.input_tokens) !== null && _a !== void 0 ? _a : 0;
-    const output = (_b = usage.output_tokens) !== null && _b !== void 0 ? _b : 0;
-    return {
-        input_tokens: input,
-        output_tokens: output,
-        total_tokens: input + output,
-    };
+    return Object.assign(Object.assign(Object.assign({ id: responseId, object: 'response', status,
+        output, model: response.model || '', created_at: Math.floor(Date.now() / 1000) }, (usage ? { usage } : {})), (incomplete_details ? { incomplete_details } : {})), { metadata: {} });
 }

package/dist/server/conversions/pairs/responses-claude/streaming.js

@@ -10,6 +10,7 @@ exports.ClaudeToResponsesConverter = void 0;
 const id_js_1 = require("../../utils/id.js");
 const stop_reasons_js_1 = require("../../utils/stop-reasons.js");
 const streaming_helpers_js_1 = require("../../utils/streaming-helpers.js");
+const usage_js_1 = require("../../utils/usage.js");
 /**
  * ClaudeToResponsesConverter: Claude Messages SSE → Responses API SSE
  */
@@ -49,7 +50,7 @@ class ClaudeToResponsesConverter {
             enumerable: true,
             configurable: true,
             writable: true,
-            value: { input_tokens: 0, output_tokens: 0 }
+            value: null
         });
         Object.defineProperty(this, "output", {
             enumerable: true,
@@ -89,7 +90,7 @@ class ClaudeToResponsesConverter {
         });
     }
     convertEvent(event) {
-        var _a, _b, _c, _d, _e;
+        var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o;
         if (!event.data)
             return [];
         const events = [];
@@ -99,6 +100,14 @@ class ClaudeToResponsesConverter {
                 case 'message_start': {
                     this.model = ((_a = data.message) === null || _a === void 0 ? void 0 : _a.model) || '';
                     this.responseId = ((_b = data.message) === null || _b === void 0 ? void 0 : _b.id) || this.responseId;
+                    // Claude 的 prompt token 数在 message_start 的 message.usage.input_tokens
+                    const startUsage = (_c = data.message) === null || _c === void 0 ? void 0 : _c.usage;
+                    if (startUsage) {
+                        this.usage = {
+                            input_tokens: (_d = startUsage.input_tokens) !== null && _d !== void 0 ? _d : 0,
+                            output_tokens: (_e = startUsage.output_tokens) !== null && _e !== void 0 ? _e : 0,
+                        };
+                    }
                     events.push(this.makeSSE('response.created', {
                         id: this.responseId,
                         object: 'response',
@@ -236,11 +245,11 @@ class ClaudeToResponsesConverter {
                     break;
                 }
                 case 'message_delta': {
-                    this.pendingStopReason = ((_c = data.delta) === null || _c === void 0 ? void 0 : _c.stop_reason) || null;
+                    this.pendingStopReason = ((_f = data.delta) === null || _f === void 0 ? void 0 : _f.stop_reason) || null;
                     if (data.usage) {
                         this.usage = {
-                            input_tokens: this.usage.input_tokens,
-                            output_tokens: (_e = (_d = data.usage.output_tokens) !== null && _d !== void 0 ? _d : data.usage.tokens) !== null && _e !== void 0 ? _e : 0,
+                            input_tokens: (_j = (_g = data.usage.input_tokens) !== null && _g !== void 0 ? _g : (_h = this.usage) === null || _h === void 0 ? void 0 : _h.input_tokens) !== null && _j !== void 0 ? _j : 0,
+                            output_tokens: (_o = (_l = (_k = data.usage.output_tokens) !== null && _k !== void 0 ? _k : data.usage.tokens) !== null && _l !== void 0 ? _l : (_m = this.usage) === null || _m === void 0 ? void 0 : _m.output_tokens) !== null && _o !== void 0 ? _o : 0,
                         };
                     }
                     break;
@@ -253,7 +262,7 @@ class ClaudeToResponsesConverter {
                     break;
             }
         }
-        catch (_f) {
+        catch (_p) {
             // Ignore parse errors
         }
         return events;
@@ -315,20 +324,9 @@ class ClaudeToResponsesConverter {
         this.closeText(events);
         this.closeThinking(events);
         const { status, incomplete_details } = (0, stop_reasons_js_1.claudeToResponsesStatus)(this.pendingStopReason);
-        const responseObj = {
-            id: this.responseId,
-            object: 'response',
-            status,
-            output: this.output,
-            model: this.model,
-            created_at: Math.floor(Date.now() / 1000),
-            usage: {
-                input_tokens: this.usage.input_tokens,
-                output_tokens: this.usage.output_tokens,
-                total_tokens: this.usage.input_tokens + this.usage.output_tokens,
-            },
-            metadata: {},
-        };
+        // 上游无 usage 时省略 usage 字段（不伪造 0，避免 Codex missing field input_tokens）
+        const responsesUsage = (0, usage_js_1.toResponsesUsage)(this.usage);
+        const responseObj = Object.assign({ id: this.responseId, object: 'response', status, output: this.output, model: this.model, created_at: Math.floor(Date.now() / 1000), metadata: {} }, (responsesUsage ? { usage: responsesUsage } : {}));
         if (incomplete_details) {
             responseObj.incomplete_details = incomplete_details;
         }

package/dist/server/conversions/pairs/responses-completions/response.js

@@ -80,15 +80,8 @@ function completionsToResponsesResponse(response) {
     const finishReason = (0, stop_reasons_js_1.completionsToResponsesFinishReason)(choice.finish_reason);
     const status = finishReason === 'incomplete' ? 'incomplete' : 'completed';
     const usage = (0, usage_js_1.completionsToResponsesUsage)(response.usage);
-    const result = {
-        id: ((_m = response.id) === null || _m === void 0 ? void 0 : _m.startsWith('resp_')) ? response.id : (0, id_js_1.generateResponseId)(),
-        object: 'response',
-        status,
-        output,
-        model: response.model,
-        usage,
-        created_at: response.created || Math.floor(Date.now() / 1000),
-    };
+    const result = Object.assign({ id: ((_m = response.id) === null || _m === void 0 ? void 0 : _m.startsWith('resp_')) ? response.id : (0, id_js_1.generateResponseId)(), object: 'response', status,
+        output, model: response.model, created_at: response.created || Math.floor(Date.now() / 1000) }, (usage ? { usage } : {}));
     if (status === 'incomplete') {
         result.incomplete_details = { reason: 'max_output_tokens' };
     }

package/dist/server/conversions/pairs/responses-completions/streaming.js

@@ -320,16 +320,9 @@ class CompletionsToResponsesConverter {
         }
         // Build full response object
         const status = this.finishReason === 'length' ? 'incomplete' : 'completed';
-        const responseObj = {
-            id: this.responseId,
-            object: 'response',
-            status,
-            output: this.output,
-            model: this.model,
-            usage: this.usage ? (0, usage_js_1.completionsToResponsesUsage)(this.usage) : {},
-            created_at: Math.floor(Date.now() / 1000),
-            metadata: {},
-        };
+        // 上游无 usage 时省略 usage 字段（不伪造 0，避免 Codex missing field input_tokens）
+        const responsesUsage = (0, usage_js_1.completionsToResponsesUsage)(this.usage);
+        const responseObj = Object.assign({ id: this.responseId, object: 'response', status, output: this.output, model: this.model, created_at: Math.floor(Date.now() / 1000), metadata: {} }, (responsesUsage ? { usage: responsesUsage } : {}));
         if (status === 'incomplete') {
             responseObj.incomplete_details = { reason: 'max_output_tokens' };
         }

package/dist/server/conversions/utils/usage.js

@@ -8,6 +8,7 @@ exports.claudeToCompletionsUsage = claudeToCompletionsUsage;
 exports.geminiToClaudeUsage = geminiToClaudeUsage;
 exports.claudeToGeminiUsage = claudeToGeminiUsage;
 exports.responsesToClaudeUsage = responsesToClaudeUsage;
+exports.toResponsesUsage = toResponsesUsage;
 exports.completionsToResponsesUsage = completionsToResponsesUsage;
 /** Map OpenAI Chat usage to Claude usage */
 function completionsToClaudeUsage(usage) {
@@ -69,14 +70,33 @@ function responsesToClaudeUsage(usage) {
         cache_creation_input_tokens: usage.cache_creation_input_tokens,
     };
 }
-/** Map OpenAI Chat usage to Responses API usage */
-function completionsToResponsesUsage(usage) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _j;
-    if (!usage)
-        return {};
+/**
+ * 构造标准 Responses API usage 对象（转换层兼容入口）。
+ *
+ * 真实值优先：input_tokens ?? prompt_tokens、output_tokens ?? completion_tokens、total_tokens。
+ * 覆盖上游 chat completions / claude / gemini 三种格式的字段命名归一。
+ *
+ * 仅当上游确实提供了任意 token 字段时返回归一化对象；否则返回 null（调用方应省略 usage 字段，
+ * 不伪造 0）。这是避免 Codex `ResponseCompleted: missing field input_tokens` 的关键——
+ * 既不吐空 `{}`，也不吐伪造的 `{0,0,0}`。
+ */
+function toResponsesUsage(usage) {
+    var _a, _b;
+    if (!usage || typeof usage !== 'object')
+        return null;
+    const input_tokens = (_a = usage.input_tokens) !== null && _a !== void 0 ? _a : usage.prompt_tokens;
+    const output_tokens = (_b = usage.output_tokens) !== null && _b !== void 0 ? _b : usage.completion_tokens;
+    const total_tokens = usage.total_tokens;
+    // 上游没返回任何 token 字段 → 不伪造，返回 null
+    if (input_tokens == null && output_tokens == null && total_tokens == null)
+        return null;
     return {
-        input_tokens: (_b = (_a = usage.input_tokens) !== null && _a !== void 0 ? _a : usage.prompt_tokens) !== null && _b !== void 0 ? _b : 0,
-        output_tokens: (_d = (_c = usage.output_tokens) !== null && _c !== void 0 ? _c : usage.completion_tokens) !== null && _d !== void 0 ? _d : 0,
-        total_tokens: (_e = usage.total_tokens) !== null && _e !== void 0 ? _e : (((_g = (_f = usage.input_tokens) !== null && _f !== void 0 ? _f : usage.prompt_tokens) !== null && _g !== void 0 ? _g : 0) + ((_j = (_h = usage.output_tokens) !== null && _h !== void 0 ? _h : usage.completion_tokens) !== null && _j !== void 0 ? _j : 0)),
+        input_tokens: input_tokens !== null && input_tokens !== void 0 ? input_tokens : 0,
+        output_tokens: output_tokens !== null && output_tokens !== void 0 ? output_tokens : 0,
+        total_tokens: total_tokens !== null && total_tokens !== void 0 ? total_tokens : ((input_tokens !== null && input_tokens !== void 0 ? input_tokens : 0) + (output_tokens !== null && output_tokens !== void 0 ? output_tokens : 0)),
     };
 }
+/** Map OpenAI Chat usage to Responses API usage（薄封装，保持现有调用方不变） */
+function completionsToResponsesUsage(usage) {
+    return toResponsesUsage(usage);
+}

package/dist/server/fs-database.js

@@ -2806,9 +2806,9 @@ class FileSystemDatabaseManager {
                 serviceStats.avgResponseTime =
                     (serviceStats.avgResponseTime * (serviceStats.totalRequests - 1) + responseTime) / serviceStats.totalRequests;
             }
-            // 更新 byModel
-            if (log.requestModel || log.targetModel) {
-                const modelName = log.requestModel || log.targetModel || 'Unknown';
+            // 更新 byModel（按实际转发的模型统计，优先 targetModel 而非客户端提交的 requestModel）
+            if (log.targetModel || log.requestModel) {
+                const modelName = log.targetModel || log.requestModel || 'Unknown';
                 let modelStats = this.statistics.byModel.find(s => s.modelName === modelName);
                 if (!modelStats) {
                     modelStats = { modelName, totalRequests: 0, totalTokens: 0, avgResponseTime: 0 };

package/dist/server/main.js

@@ -24,6 +24,7 @@ const proxy_server_1 = require("./proxy-server");
 const index_1 = require("./access-keys/index");
 const manager_1 = require("./access-keys/manager");
 const policy_manager_1 = require("./access-keys/policy-manager");
+const performance_tracker_1 = require("./performance-tracker");
 const os_1 = __importDefault(require("os"));
 const auth_1 = require("./auth");
 const version_check_1 = require("./version-check");
@@ -383,7 +384,7 @@ const DEFAULT_CODEX_REASONING_EFFORT = 'high';
 const isCodexReasoningEffort = (value) => {
     return typeof value === 'string' && VALID_CODEX_REASONING_EFFORTS.includes(value);
 };
-const writeCodexConfig = (_dbManager_1, ...args_1) => __awaiter(void 0, [_dbManager_1, ...args_1], void 0, function* (_dbManager, modelReasoningEffort = DEFAULT_CODEX_REASONING_EFFORT, codexDefaultModel, options = {}) {
+const writeCodexConfig = (_dbManager_1, ...args_1) => __awaiter(void 0, [_dbManager_1, ...args_1], void 0, function* (_dbManager, modelReasoningEffort = DEFAULT_CODEX_REASONING_EFFORT, codexDefaultModel, enableMemories, options = {}) {
     var _a;
     try {
         const homeDir = os_1.default.homedir();
@@ -450,6 +451,17 @@ const writeCodexConfig = (_dbManager_1, ...args_1) => __awaiter(void 0, [_dbMana
                 }
             }
         };
+        // 记忆功能配置
+        if (enableMemories) {
+            proxyConfig.features = {
+                memories: true,
+            };
+            proxyConfig.memories = {
+                generate_memories: true,
+                use_memories: true,
+                disable_on_external_context: true,
+            };
+        }
         // 使用智能合并
         const mergedConfig = (0, config_merge_1.mergeTomlConfig)(proxyConfig, currentConfig, config_managed_fields_1.CODEX_CONFIG_MANAGED_FIELDS);
         // 原子性写入合并后的配置
@@ -681,7 +693,7 @@ const syncConfigsOnServerStartup = (dbManager) => __awaiter(void 0, void 0, void
     const modelReasoningEffort = isCodexReasoningEffort(config.codexModelReasoningEffort)
         ? config.codexModelReasoningEffort
         : DEFAULT_CODEX_REASONING_EFFORT;
-    const codexWritten = yield writeCodexConfig(dbManager, modelReasoningEffort, config.codexDefaultModel);
+    const codexWritten = yield writeCodexConfig(dbManager, modelReasoningEffort, config.codexDefaultModel, config.codexEnableMemories);
     console.log(`[Startup Config Sync] Codex config ${codexWritten ? 'written' : 'skipped'}`);
 });
 const syncConfigsOnGlobalConfigUpdate = (dbManager) => __awaiter(void 0, void 0, void 0, function* () {
@@ -694,7 +706,7 @@ const syncConfigsOnGlobalConfigUpdate = (dbManager) => __awaiter(void 0, void 0,
     const modelReasoningEffort = isCodexReasoningEffort(config.codexModelReasoningEffort)
         ? config.codexModelReasoningEffort
         : DEFAULT_CODEX_REASONING_EFFORT;
-    const codexUpdated = yield writeCodexConfig(dbManager, modelReasoningEffort, config.codexDefaultModel, { allowOverwriteRefresh: true });
+    const codexUpdated = yield writeCodexConfig(dbManager, modelReasoningEffort, config.codexDefaultModel, config.codexEnableMemories, { allowOverwriteRefresh: true });
     console.log(`[Config Update Sync] Codex config ${codexUpdated ? 'written' : 'skipped'}`);
 });
 const getCentralSkillsDir = () => {
@@ -2038,7 +2050,11 @@ ${instruction}
             : isCodexReasoningEffort(appConfig.codexModelReasoningEffort)
                 ? appConfig.codexModelReasoningEffort
                 : DEFAULT_CODEX_REASONING_EFFORT;
-        const result = yield writeCodexConfig(dbManager, modelReasoningEffort, appConfig.codexDefaultModel);
+        const requestedEnableMemories = req.body.enableMemories;
+        const enableMemories = requestedEnableMemories !== undefined
+            ? !!requestedEnableMemories
+            : !!appConfig.codexEnableMemories;
+        const result = yield writeCodexConfig(dbManager, modelReasoningEffort, appConfig.codexDefaultModel, enableMemories);
         applyWriteLocalRecords(proxyServer);
         res.json(result);
     })));
@@ -3073,6 +3089,64 @@ ${instruction}
         }
         res.json(alerts);
     })));
+    // ============ 服务性能统计（全局，与 AUTH 无关） ============
+    const requirePerfTracker = (res) => {
+        const tracker = proxyServer.getPerformanceTracker();
+        if (!tracker) {
+            res.status(503).json({ error: 'Performance tracker not initialized' });
+        }
+        return tracker;
+    };
+    // 全部 API 服务平铺一览（含所属供应商）
+    app.get('/api/performance/services-overview', asyncHandler((_req, res) => __awaiter(void 0, void 0, void 0, function* () {
+        const tracker = requirePerfTracker(res);
+        if (!tracker)
+            return;
+        res.json(tracker.getServicesOverview());
+    })));
+    // 全部供应商一览
+    app.get('/api/performance/vendors', asyncHandler((_req, res) => __awaiter(void 0, void 0, void 0, function* () {
+        const tracker = requirePerfTracker(res);
+        if (!tracker)
+            return;
+        res.json(tracker.getVendorsOverview());
+    })));
+    // 某供应商详情（rollup + 其下服务）
+    app.get('/api/performance/vendors/:vendorId', asyncHandler((req, res) => __awaiter(void 0, void 0, void 0, function* () {
+        const tracker = requirePerfTracker(res);
+        if (!tracker)
+            return;
+        const detail = tracker.getVendorDetail(req.params.vendorId);
+        if (!detail) {
+            res.status(404).json({ error: 'Vendor not found' });
+            return;
+        }
+        res.json(detail);
+    })));
+    // 某服务详情（rollup + 其下模型）
+    app.get('/api/performance/services/:serviceId', asyncHandler((req, res) => __awaiter(void 0, void 0, void 0, function* () {
+        const tracker = requirePerfTracker(res);
+        if (!tracker)
+            return;
+        const detail = tracker.getServiceDetail(req.params.serviceId);
+        if (!detail) {
+            res.status(404).json({ error: 'Service not found' });
+            return;
+        }
+        res.json(detail);
+    })));
+    // 单模型详情（派生 + 小时走势 + 极值）
+    app.get('/api/performance/services/:serviceId/models/:model', asyncHandler((req, res) => __awaiter(void 0, void 0, void 0, function* () {
+        const tracker = requirePerfTracker(res);
+        if (!tracker)
+            return;
+        const detail = tracker.getModelDetail(req.params.serviceId, decodeURIComponent(req.params.model));
+        if (!detail) {
+            res.status(404).json({ error: 'Model not found' });
+            return;
+        }
+        res.json(detail);
+    })));
     // 写入MCP配置到Claude Code或Codex的全局配置文件
     const writeMCPConfig = (targetType) => __awaiter(void 0, void 0, void 0, function* () {
         try {
@@ -3320,6 +3394,16 @@ const start = () => __awaiter(void 0, void 0, void 0, function* () {
     catch (error) {
         console.error('[Server] AccessKey module initialization failed:', error);
     }
+    // Initialize Service Performance Tracker (全局统计，与 AUTH 无关)
+    const performanceTracker = new performance_tracker_1.ServicePerformanceTracker(dataDir);
+    try {
+        yield performanceTracker.initialize();
+        performanceTracker.startAutoFlush();
+        proxyServer.setPerformanceTracker(performanceTracker);
+    }
+    catch (error) {
+        console.error('[Server] Performance tracker initialization failed:', error);
+    }
     // 恢复已写入本地的 AccessKey（在代理配置写入之后、AccessKey 模块初始化之后）
     try {
         applyWriteLocalRecords(proxyServer);
@@ -3424,6 +3508,14 @@ const start = () => __awaiter(void 0, void 0, void 0, function* () {
             catch (error) {
                 console.error('[Shutdown ...] AccessKey module shutdown failed:', error);
             }
+            // Flush 服务性能统计（全局桶）后停止定时刷盘
+            try {
+                performanceTracker.stopAutoFlush();
+                yield performanceTracker.flush();
+            }
+            catch (error) {
+                console.error('[Shutdown ...] Performance tracker flush failed:', error);
+            }
             dbManager.close();
             // 清理规则状态广播器（关闭 SSE 连接）
             rules_status_service_1.rulesStatusBroadcaster.destroy();