alemonjs-aichat 1.0.34 → 1.0.37

package/lib/config.js

@@ -3,6 +3,8 @@ import fs from 'fs';
 import path from 'path';
 import { redis } from './redis.js';
 
+const TOOL_LOOP_TTL_SECONDS = 600;
+const TOOL_LOOP_STALE_MS = TOOL_LOOP_TTL_SECONDS * 1000;
 class db {
     redis;
     systemPrompt = "";
@@ -427,9 +429,50 @@ class db {
             for (const key of keys) {
                 await this.redis.del(key);
             }
+            const contextStatsKeys = await this.redis.keys(`ai:context_stats:*`);
+            if (contextStatsKeys.length > 0) {
+                await this.redis.del(...contextStatsKeys);
+            }
+            await this.clearAITransientState();
             return;
         }
-        await this.redis.del(`ai:history:${guid}`);
+        await this.clearAITransientState(guid);
+        await this.redis.del(`ai:history:${guid}`, `ai:context_stats:${guid}`);
+    }
+    /** 记录当前会话上下文和 token 使用情况 */
+    async recordAIContextStats(guid, input) {
+        const current = await this.getAIContextStats(guid);
+        const promptTokens = input.promptTokens ?? 0;
+        const completionTokens = input.completionTokens ?? 0;
+        const totalTokens = input.totalTokens ?? promptTokens + completionTokens;
+        const cachedTokens = input.cachedTokens ?? 0;
+        const cacheMissTokens = input.cacheMissTokens ?? Math.max(promptTokens - cachedTokens, 0);
+        const nextStats = {
+            ...input,
+            promptTokens: input.promptTokens ?? null,
+            completionTokens: input.completionTokens ?? null,
+            totalTokens: input.totalTokens ?? null,
+            cachedTokens: input.cachedTokens ?? null,
+            cacheMissTokens: input.cacheMissTokens ?? null,
+            guid,
+            requestCount: (current?.requestCount ?? 0) + 1,
+            updatedAt: Date.now(),
+            totalPromptTokens: (current?.totalPromptTokens ?? 0) + promptTokens,
+            totalCompletionTokens: (current?.totalCompletionTokens ?? 0) + completionTokens,
+            totalTokensUsed: (current?.totalTokensUsed ?? 0) + totalTokens,
+            totalCachedTokens: (current?.totalCachedTokens ?? 0) + cachedTokens,
+            totalCacheMissTokens: (current?.totalCacheMissTokens ?? 0) + cacheMissTokens,
+        };
+        await this.redis.set(`ai:context_stats:${guid}`, JSON.stringify(nextStats));
+        return nextStats;
+    }
+    /** 获取当前会话上下文和 token 使用情况 */
+    async getAIContextStats(guid) {
+        const statsStr = await this.redis.get(`ai:context_stats:${guid}`);
+        if (!statsStr) {
+            return null;
+        }
+        return JSON.parse(statsStr);
     }
     /** 清理最后的N条聊天记录 */
     async clearLastNChatHistory(guid, n) {
@@ -464,17 +507,72 @@ class db {
         await this.redis.set(this.getArchiveRecordKey(guid, id), JSON.stringify(record));
         return record;
     }
+    /** 清理工具循环等临时会话状态 */
+    async clearAITransientState(guid) {
+        if (!guid) {
+            const transientPrefixes = [
+                "ai:tool_loop_processing:",
+                "ai:guidance:",
+                "ai:session:",
+            ];
+            for (const prefix of transientPrefixes) {
+                const keys = await this.redis.keys(`${prefix}*`);
+                if (keys.length > 0) {
+                    await this.redis.del(...keys);
+                }
+            }
+            return;
+        }
+        await this.redis.del(`ai:tool_loop_processing:${guid}`, `ai:guidance:${guid}`, `ai:session:${guid}`);
+    }
     /** 设置工具调用中的处理状态 */
     async setAIToolLoopProcessing(guid, processing) {
         if (processing) {
-            await this.redis.set(`ai:tool_loop_processing:${guid}`, "1");
+            const now = Date.now();
+            const state = {
+                processing: true,
+                startedAt: now,
+                updatedAt: now,
+            };
+            await this.redis.set(`ai:tool_loop_processing:${guid}`, JSON.stringify(state), "EX", TOOL_LOOP_TTL_SECONDS);
             return;
         }
         await this.redis.del(`ai:tool_loop_processing:${guid}`);
     }
     /** 获取工具调用中的处理状态 */
     async getAIToolLoopProcessing(guid) {
-        return (await this.redis.get(`ai:tool_loop_processing:${guid}`)) || "0";
+        const key = `ai:tool_loop_processing:${guid}`;
+        const processing = await this.redis.get(key);
+        if (!processing) {
+            return "0";
+        }
+        let state = null;
+        if (processing !== "1") {
+            try {
+                const parsed = JSON.parse(processing);
+                if (parsed?.processing === true) {
+                    state = parsed;
+                }
+            }
+            catch {
+                await this.clearAITransientState(guid);
+                return "0";
+            }
+            const updatedAt = Number(state?.updatedAt || state?.startedAt || 0);
+            if (!updatedAt || Date.now() - updatedAt > TOOL_LOOP_STALE_MS) {
+                await this.clearAITransientState(guid);
+                return "0";
+            }
+        }
+        const ttl = await this.redis.ttl(key);
+        if (ttl === -1) {
+            if (processing === "1") {
+                await this.clearAITransientState(guid);
+                return "0";
+            }
+            await this.redis.expire(key, TOOL_LOOP_TTL_SECONDS);
+        }
+        return "1";
     }
     /** 获取工具调用期间追加的引导消息 */
     async getAIGuidanceMessages(guid) {
@@ -482,13 +580,20 @@ class db {
         if (!queueStr) {
             return [];
         }
-        return JSON.parse(queueStr);
+        try {
+            const queue = JSON.parse(queueStr);
+            return Array.isArray(queue) ? queue : [];
+        }
+        catch {
+            await this.clearAIGuidanceMessages(guid);
+            return [];
+        }
     }
     /** 追加工具调用期间的引导消息 */
     async addAIGuidanceMessage(guid, data) {
         const queue = await this.getAIGuidanceMessages(guid);
         queue.push(data);
-        await this.redis.set(`ai:guidance:${guid}`, JSON.stringify(queue));
+        await this.redis.set(`ai:guidance:${guid}`, JSON.stringify(queue), "EX", TOOL_LOOP_TTL_SECONDS);
         return queue;
     }
     /** 提取并清空工具调用期间的引导消息 */

package/lib/response/config/res.js

@@ -1,7 +1,7 @@
 import redisClient from '../../config.js';
 import { loadSkills } from '../../api/loadSkill.js';
 import App from '../../image/conponent/AiConfig.js';
-import { useMessage, Image, Text } from 'alemonjs';
+import { useMessage, Text, Image } from 'alemonjs';
 import { renderComponentIsHtmlToBuffer } from 'jsxp';
 import OpenAi from 'openai';
 
@@ -54,10 +54,58 @@ const archiveCurrentConversation = async (guid, reason, aiConfig) => {
         reason,
     });
 };
+const formatTokenValue = (value) => {
+    return typeof value === "number" && Number.isFinite(value)
+        ? String(value)
+        : "未知";
+};
+const formatAIContextStatus = (stats) => {
+    if (!stats) {
+        return "当前会话还没有 CAPI 上下文统计。";
+    }
+    const updatedAt = new Date(stats.updatedAt).toLocaleString("zh-CN", {
+        hour12: false,
+    });
+    const usageNote = stats.usageAvailable
+        ? ""
+        : "\n提示: 最近一次请求未返回 usage, 实际 token 统计可能不完整。";
+    return [
+        "当前AI状态：",
+        `模式: ${stats.mode.toUpperCase()}`,
+        `模型: ${stats.model || "未知"}`,
+        `更新时间: ${updatedAt}`,
+        `请求次数: ${stats.requestCount}`,
+        "",
+        "最近一次上下文：",
+        `消息数: ${stats.contextMessageCount}`,
+        `上下文字符数: ${stats.contextCharLength}`,
+        `上下文估算token: ${stats.estimatedContextTokens}`,
+        `输入token: ${formatTokenValue(stats.promptTokens)}`,
+        `输出token: ${formatTokenValue(stats.completionTokens)}`,
+        `总token: ${formatTokenValue(stats.totalTokens)}`,
+        `缓存命中token: ${formatTokenValue(stats.cachedTokens)}`,
+        `缓存未命中token: ${formatTokenValue(stats.cacheMissTokens)}`,
+        "",
+        "当前会话累计：",
+        `输入token: ${stats.totalPromptTokens}`,
+        `输出token: ${stats.totalCompletionTokens}`,
+        `总token: ${stats.totalTokensUsed}`,
+        `缓存命中token: ${stats.totalCachedTokens}`,
+        `缓存未命中token: ${stats.totalCacheMissTokens}`,
+        usageNote,
+    ]
+        .filter((line) => line !== "")
+        .join("\n");
+};
 var res = onResponse(selects, async (e, next) => {
     // 创建
     const [message] = useMessage(e);
     const config = await redisClient.getAIConfig(e.guid);
+    // 查看AI状态
+    if (/^(\/|#)ai状态$/i.test(e.msg)) {
+        const stats = await redisClient.getAIContextStats(e.guid);
+        message.send(format(Text(formatAIContextStatus(stats))));
+    }
     // 查看AI配置
     if (/^(\/|#)(ai配置|当前提示词|查看提示词)$/i.test(e.msg)) {
         const complexResponse = (await redisClient.getComplexOutput(e.guid)) || "1"; // 复杂输出

package/lib/response/mw.js

@@ -1,9 +1,9 @@
 var mw = onMiddleware(onSelects(["message.create", "private.message.create"]), async (event, next) => {
     if (event.Xianyu) {
-        return true; // 继续执行后续的处理函数
+        return next(); // 普通聊天交给后续路由/文件处理，跳过命令子路由
     }
     else {
-        return next(); // 跳过所有指令,继续执行后续的处理函数
+        return true; // 命令消息继续执行当前命令子路由
     }
 });
 

package/lib/response/zreply/capi.js

@@ -43,6 +43,56 @@ const appendPendingGuidanceMessages = async (guid, messages) => {
     await redisClient.addAIChatHistoryBatch(guid, guidanceMessages);
     return guidanceMessages;
 };
+const getTextLength = (value) => {
+    if (typeof value === "string") {
+        return value.length;
+    }
+    if (Array.isArray(value)) {
+        return value.reduce((sum, item) => sum + getTextLength(item), 0);
+    }
+    if (value && typeof value === "object") {
+        return Object.values(value).reduce((sum, item) => sum + getTextLength(item), 0);
+    }
+    return 0;
+};
+const getNumber = (value) => {
+    return typeof value === "number" && Number.isFinite(value) ? value : null;
+};
+const extractCachedTokens = (usage) => {
+    return getNumber(usage?.prompt_tokens_details?.cached_tokens ??
+        usage?.input_tokens_details?.cached_tokens ??
+        usage?.prompt_cache_hit_tokens);
+};
+const extractCacheMissTokens = (usage, promptTokens) => {
+    const explicitMissTokens = getNumber(usage?.prompt_cache_miss_tokens);
+    if (explicitMissTokens !== null) {
+        return explicitMissTokens;
+    }
+    const cachedTokens = extractCachedTokens(usage);
+    if (promptTokens !== null && cachedTokens !== null) {
+        return Math.max(promptTokens - cachedTokens, 0);
+    }
+    return null;
+};
+const recordCapiContextStats = async (guid, model, messages, completion) => {
+    const usage = completion?.usage;
+    const promptTokens = getNumber(usage?.prompt_tokens ?? usage?.input_tokens);
+    const completionTokens = getNumber(usage?.completion_tokens ?? usage?.output_tokens);
+    const contextCharLength = getTextLength(messages);
+    await redisClient.recordAIContextStats(guid, {
+        mode: "capi",
+        model,
+        contextMessageCount: messages.length,
+        contextCharLength,
+        estimatedContextTokens: Math.ceil(contextCharLength / 4),
+        promptTokens,
+        completionTokens,
+        totalTokens: getNumber(usage?.total_tokens),
+        cachedTokens: extractCachedTokens(usage),
+        cacheMissTokens: extractCacheMissTokens(usage, promptTokens),
+        usageAvailable: Boolean(usage),
+    });
+};
 const CApiReply = async (e) => {
     console.log("e.UserId", e.UserId, e.bot);
     const cfg = await getChatConfig(e);
@@ -81,6 +131,7 @@ const CApiReply = async (e) => {
         createParams["stream"] = false;
         // log("请求AI，参数：", createParams);
         const stream = (await openai.chat.completions.create(createParams));
+        await recordCapiContextStats(e.guid, cfg.config.model, messages, stream);
         log("AI回复原始数据:\n", JSON.stringify(stream, null, 2));
         let fullContent = "";
         let toolCalls = [];
@@ -140,6 +191,7 @@ const CApiReply = async (e) => {
                     }
                     Object.assign(params, getDeepThoughtReasoning(cfg));
                     res = await openai.chat.completions.create(params);
+                    await recordCapiContextStats(e.guid, cfg.config.model, messages, res);
                     if (!res.choices || res.choices.length === 0) {
                         log("AI未返回内容");
                         return;
@@ -243,6 +295,7 @@ const CApiReply = async (e) => {
                 // log("重新请求AI，参数：", params);
                 // 重新请求
                 res = await openai.chat.completions.create(params);
+                await recordCapiContextStats(e.guid, cfg.config.model, messages, res);
                 // 检查是否还有工具调用需要处理
                 if (!res.choices || res.choices.length === 0) {
                     log("AI未返回内容");
@@ -273,6 +326,7 @@ const CApiReply = async (e) => {
         };
     }
     catch (error) {
+        await redisClient.setAIToolLoopProcessing(e.guid, false);
         console.error("AI回复出错", error);
         message.send(format(Text("AI回复出错了\n" + error)));
         return {

package/lib/routes/commands.js

@@ -14,7 +14,7 @@ const children = [
     },
     // config
     {
-        regular: Regular.or(/^(\/|#)(ai配置|当前提示词|查看提示词)$/i, /^(\/|#)ai列表$/i, /^(\/|#)提示词列表$/i, /^(\/|#)清空对话$/i, /^(\/|#)清空(所有|全部)对话$/i, /^(\/|#)get (.+)$/i, /(\/|#)(开启)?(new|新对话|新会话)$/i),
+        regular: Regular.or(/^(\/|#)(ai配置|当前提示词|查看提示词)$/i, /^(\/|#)ai列表$/i, /^(\/|#)提示词列表$/i, /^(\/|#)清空对话$/i, /^(\/|#)清空(所有|全部)对话$/i, /^(\/|#)get (.+)$/i, /(\/|#)(开启)?(new|新对话|新会话)$/i, /^(\/|#)ai状态$/),
         handler: lazy(() => import('../response/config/res.js')),
     },
     // setting (包含大量子命令)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "alemonjs-aichat",
-  "version": "1.0.34",
+  "version": "1.0.37",
   "description": "alemonjs-aichat",
   "author": "suancaixianyu",
   "license": "MIT",