aicodeswitch 3.9.2 → 3.9.3

package/dist/server/proxy-server.js

@@ -887,7 +887,8 @@ class ProxyServer {
                 return undefined;
             const body = req.body;
             const requestModel = body === null || body === void 0 ? void 0 : body.model;
-            const contentType = forcedContentType || this.determineContentType(req);
+            const route = this.dbManager.getRoutes().find(r => r.id === routeId);
+            const contentType = forcedContentType || this.determineContentType(req, (route === null || route === void 0 ? void 0 : route.targetType) || 'claude-code', routeId);
             // 高智商规则优先于 model-mapping，确保 !!/推断命中时不会被模型映射覆盖
             if (contentType === 'high-iq') {
                 const highIqRules = enabledRules.filter(rule => rule.contentType === 'high-iq');
@@ -910,7 +911,32 @@ class ProxyServer {
                     return rule;
                 }
             }
-            // 1. 首先查找 model-mapping 类型的规则，按 sortOrder 降序匹配
+            // 1. 查找其他内容类型的规则
+            const contentTypeRules = enabledRules.filter(rule => rule.contentType === contentType);
+            // 过滤黑名单和token限制
+            for (const rule of contentTypeRules) {
+                const isBlacklisted = yield this.dbManager.isServiceBlacklisted(rule.targetServiceId, routeId, contentType);
+                if (isBlacklisted) {
+                    continue;
+                }
+                // 检查并重置到期的规则
+                this.dbManager.checkAndResetRuleIfNeeded(rule.id);
+                this.dbManager.checkAndResetRequestCountIfNeeded(rule.id);
+                // 检查token限制（tokenLimit单位是k，需要乘以1000转换为实际token数）
+                if (rule.tokenLimit && rule.totalTokensUsed !== undefined && rule.totalTokensUsed >= rule.tokenLimit * 1000) {
+                    continue; // 跳过超限规则
+                }
+                // 检查请求次数限制
+                if (rule.requestCountLimit && rule.totalRequestsUsed !== undefined && rule.totalRequestsUsed >= rule.requestCountLimit) {
+                    continue; // 跳过超限规则
+                }
+                // 检查频率限制
+                if (this.isFrequencyLimitExceeded(rule)) {
+                    continue; // 跳过达到频率限制的规则
+                }
+                return rule;
+            }
+            // 2. 然后查找 model-mapping 类型的规则
             if (requestModel) {
                 const modelMappingRules = enabledRules.filter(rule => rule.contentType === 'model-mapping' &&
                     rule.replacedModel &&
@@ -939,31 +965,6 @@ class ProxyServer {
                     return rule;
                 }
             }
-            // 2. 查找其他内容类型的规则
-            const contentTypeRules = enabledRules.filter(rule => rule.contentType === contentType);
-            // 过滤黑名单和token限制
-            for (const rule of contentTypeRules) {
-                const isBlacklisted = yield this.dbManager.isServiceBlacklisted(rule.targetServiceId, routeId, contentType);
-                if (isBlacklisted) {
-                    continue;
-                }
-                // 检查并重置到期的规则
-                this.dbManager.checkAndResetRuleIfNeeded(rule.id);
-                this.dbManager.checkAndResetRequestCountIfNeeded(rule.id);
-                // 检查token限制（tokenLimit单位是k，需要乘以1000转换为实际token数）
-                if (rule.tokenLimit && rule.totalTokensUsed !== undefined && rule.totalTokensUsed >= rule.tokenLimit * 1000) {
-                    continue; // 跳过超限规则
-                }
-                // 检查请求次数限制
-                if (rule.requestCountLimit && rule.totalRequestsUsed !== undefined && rule.totalRequestsUsed >= rule.requestCountLimit) {
-                    continue; // 跳过超限规则
-                }
-                // 检查频率限制
-                if (this.isFrequencyLimitExceeded(rule)) {
-                    continue; // 跳过达到频率限制的规则
-                }
-                return rule;
-            }
             // 3. 最后返回 default 规则
             const defaultRules = enabledRules.filter(rule => rule.contentType === 'default');
             // 过滤黑名单和token限制
@@ -1003,7 +1004,8 @@ class ProxyServer {
         const body = req.body;
         const requestModel = body === null || body === void 0 ? void 0 : body.model;
         const candidates = [];
-        const contentType = forcedContentType || this.determineContentType(req);
+        const route = this.dbManager.getRoutes().find(r => r.id === routeId);
+        const contentType = forcedContentType || this.determineContentType(req, (route === null || route === void 0 ? void 0 : route.targetType) || 'claude-code', routeId);
         const prioritizeContentType = contentType === 'high-iq';
         const modelMappingRules = requestModel
             ? enabledRules.filter(rule => rule.contentType === 'model-mapping' &&
@@ -1133,7 +1135,7 @@ class ProxyServer {
             }
         }
     }
-    determineContentType(req) {
+    determineContentType(req, targetType, routeId) {
         const body = req.body;
         if (!body)
             return 'default';
@@ -1145,8 +1147,10 @@ class ProxyServer {
         if (explicitType) {
             return explicitType;
         }
+        // 获取sessionId用于session级别的检测（如long-context）
+        const sessionId = this.defaultExtractSessionId(req, targetType);
         for (const detector of this.getContentTypeDetectors()) {
-            if (detector.match(req, body)) {
+            if (detector.match(req, body, sessionId, routeId)) {
                 return detector.type;
             }
         }
@@ -1158,17 +1162,17 @@ class ProxyServer {
                 type: 'image-understanding',
                 match: (_req, body) => this.containsImageContent(body.messages) || this.containsImageContent(body.input),
             },
-            {
-                type: 'thinking',
-                match: (_req, body) => this.hasThinkingSignal(body),
-            },
             {
                 type: 'high-iq',
                 match: (_req, body) => this.hasHighIqSignal(body),
             },
             {
                 type: 'long-context',
-                match: (_req, body) => this.hasLongContextSignal(body),
+                match: (_req, body, sessionId, routeId) => this.hasLongContextSignal(body, sessionId, routeId),
+            },
+            {
+                type: 'thinking',
+                match: (_req, body) => this.hasThinkingSignal(body),
             },
             {
                 type: 'background',
@@ -1574,8 +1578,8 @@ class ProxyServer {
         ];
         return candidates.some((value) => value === true || value === 'background');
     }
-    hasLongContextSignal(body) {
-        var _a, _b;
+    hasLongContextSignal(body, sessionId, routeId) {
+        var _a, _b, _c;
         const explicit = [
             body === null || body === void 0 ? void 0 : body.long_context,
             body === null || body === void 0 ? void 0 : body.longContext,
@@ -1585,6 +1589,22 @@ class ProxyServer {
         if (explicit.some((value) => value === true)) {
             return true;
         }
+        // 检查session累积tokens
+        if (sessionId && routeId) {
+            const session = this.dbManager.getSession(sessionId);
+            if (session && session.totalTokens > 0) {
+                // 查找该route下的long-context规则，获取阈值配置
+                const rules = this.getRulesByRouteId(routeId);
+                const longContextRule = rules === null || rules === void 0 ? void 0 : rules.find(rule => rule.contentType === 'long-context' && !rule.isDisabled);
+                // 默认阈值为1M tokens (1000k)
+                const defaultThreshold = 1000; // 单位：k
+                const threshold = (_c = longContextRule === null || longContextRule === void 0 ? void 0 : longContextRule.sessionTokenThreshold) !== null && _c !== void 0 ? _c : defaultThreshold;
+                // 如果session累积tokens超过阈值，则认为是long-context
+                if (session.totalTokens >= threshold * 1000) {
+                    return true;
+                }
+            }
+        }
         const maxTokens = this.extractNumericField(body, [
             'max_tokens',
             'max_output_tokens',
@@ -2273,8 +2293,9 @@ class ProxyServer {
                 // Session 索引逻辑
                 const sessionId = this.defaultExtractSessionId(req, targetType);
                 if (sessionId) {
-                    const totalTokens = ((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.inputTokens) || 0) + ((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.outputTokens) || 0) +
-                        ((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.totalTokens) || 0);
+                    // 正确计算当前请求的tokens：优先使用totalTokens，否则使用input+output
+                    const totalTokens = (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));
                     const sessionTitle = this.defaultExtractSessionTitle(req, sessionId);
                     const existingSession = this.dbManager.getSession(sessionId);
                     this.dbManager.upsertSession({
@@ -2298,7 +2319,8 @@ class ProxyServer {
                 }
                 // 更新规则的token使用量（只在成功请求时更新）
                 if (usageForLog && statusCode < 400) {
-                    const totalTokens = (usageForLog.inputTokens || 0) + (usageForLog.outputTokens || 0);
+                    const totalTokens = usageForLog.totalTokens ||
+                        ((usageForLog.inputTokens || 0) + (usageForLog.outputTokens || 0));
                     if (totalTokens > 0) {
                         this.dbManager.incrementRuleTokenUsage(rule.id, totalTokens);
                         // 获取更新后的规则数据并广播