chattercatcher 0.1.16 → 0.1.17

package/dist/cli.js

@@ -8,7 +8,7 @@ import fs13 from "fs/promises";
 // package.json
 var package_default = {
   name: "chattercatcher",
-  version: "0.1.16",
+  version: "0.1.17",
   description: "\u672C\u5730\u4F18\u5148\u7684\u98DE\u4E66/Lark \u5BB6\u5EAD\u7FA4\u77E5\u8BC6\u5E93\u673A\u5668\u4EBA",
   type: "module",
   main: "dist/index.js",
@@ -894,6 +894,40 @@ function getGatewayStatus(config, secrets) {
 function normalizeBaseUrl(baseUrl) {
   return baseUrl.replace(/\/+$/, "");
 }
+function toOpenAIMessage(message) {
+  return {
+    role: message.role,
+    content: message.content,
+    ...message.toolCallId ? { tool_call_id: message.toolCallId } : {},
+    ...message.toolCalls ? {
+      tool_calls: message.toolCalls.map((toolCall) => ({
+        id: toolCall.id,
+        type: "function",
+        function: {
+          name: toolCall.name,
+          arguments: JSON.stringify(toolCall.input)
+        }
+      }))
+    } : {}
+  };
+}
+function toOpenAITool(tool) {
+  return {
+    type: "function",
+    function: {
+      name: tool.name,
+      description: tool.description,
+      parameters: tool.inputSchema
+    }
+  };
+}
+function parseToolCalls(message) {
+  return message?.tool_calls?.map((toolCall) => ({
+    id: toolCall.id,
+    name: toolCall.function.name,
+    input: JSON.parse(toolCall.function.arguments)
+  })) ?? [];
+}
 var OpenAICompatibleChatModel = class {
   constructor(options) {
     this.options = options;
@@ -911,7 +945,7 @@ var OpenAICompatibleChatModel = class {
       },
       body: JSON.stringify({
         model: this.options.model,
-        messages,
+        messages: messages.map(toOpenAIMessage),
         temperature: this.options.temperature ?? 0.2
       })
     });
@@ -926,6 +960,35 @@ var OpenAICompatibleChatModel = class {
     }
     return content;
   }
+  async completeWithTools(messages, tools) {
+    if (!this.options.baseUrl || !this.options.apiKey || !this.options.model) {
+      throw new Error("LLM \u914D\u7F6E\u4E0D\u5B8C\u6574\u3002\u8BF7\u8FD0\u884C chattercatcher setup \u6216 chattercatcher settings\u3002");
+    }
+    const response = await fetch(`${normalizeBaseUrl(this.options.baseUrl)}/chat/completions`, {
+      method: "POST",
+      headers: {
+        authorization: `Bearer ${this.options.apiKey}`,
+        "content-type": "application/json"
+      },
+      body: JSON.stringify({
+        model: this.options.model,
+        messages: messages.map(toOpenAIMessage),
+        tools: tools.map(toOpenAITool),
+        tool_choice: "auto",
+        temperature: this.options.temperature ?? 0.2
+      })
+    });
+    if (!response.ok) {
+      const body = await response.text();
+      throw new Error(`LLM \u8BF7\u6C42\u5931\u8D25\uFF1A${response.status} ${body}`);
+    }
+    const data2 = await response.json();
+    const message = data2.choices?.[0]?.message;
+    return {
+      content: message?.content ?? "",
+      toolCalls: parseToolCalls(message)
+    };
+  }
 };
 var OpenAICompatibleEmbeddingModel = class {
   constructor(options) {
@@ -1612,6 +1675,73 @@ var MessageFtsRetriever = class {
   }
 };
 
+// src/rag/search-tools.ts
+var searchInputSchema = {
+  type: "object",
+  properties: {
+    query: { type: "string", description: "Search query written by the model." },
+    limit: { type: "number", description: "Maximum number of evidence blocks to return." }
+  },
+  required: ["query"],
+  additionalProperties: false
+};
+function parseSearchInput(input2) {
+  const rawQuery = typeof input2 === "object" && input2 !== null && "query" in input2 ? input2.query : void 0;
+  if (typeof rawQuery !== "string") {
+    throw new Error("\u641C\u7D22 query \u5FC5\u987B\u662F\u975E\u7A7A\u5B57\u7B26\u4E32\u3002");
+  }
+  const query = rawQuery.trim();
+  if (!query) {
+    throw new Error("\u641C\u7D22 query \u5FC5\u987B\u662F\u975E\u7A7A\u5B57\u7B26\u4E32\u3002");
+  }
+  const rawLimit = typeof input2 === "object" && input2 !== null && "limit" in input2 ? input2.limit : void 0;
+  const numericLimit = typeof rawLimit === "number" && Number.isFinite(rawLimit) ? rawLimit : 5;
+  const limit = Math.min(12, Math.max(1, Math.floor(numericLimit)));
+  return { query, limit };
+}
+async function runRetriever(retriever, input2) {
+  const { query, limit } = parseSearchInput(input2);
+  const results = await retriever.retrieve(query);
+  return results.slice(0, limit);
+}
+function createSearchTool(name, description, retriever) {
+  return {
+    name,
+    description,
+    inputSchema: searchInputSchema,
+    execute: (input2) => runRetriever(retriever, input2)
+  };
+}
+function createRagSearchTools(input2) {
+  const tools = [
+    createSearchTool(
+      "hybrid_search",
+      "Search across all indexed RAG evidence using the default hybrid retrieval strategy.",
+      input2.hybrid
+    ),
+    createSearchTool(
+      "search_messages",
+      "Search chat messages only when the answer likely depends on message-level evidence.",
+      input2.messages
+    ),
+    createSearchTool(
+      "search_episodes",
+      "Search episode summaries only when the answer likely depends on longer-running context.",
+      input2.episodes
+    )
+  ];
+  if (input2.semantic) {
+    tools.push(
+      createSearchTool(
+        "semantic_search",
+        "Search semantic vector evidence only when broader conceptual recall is needed.",
+        input2.semantic
+      )
+    );
+  }
+  return tools;
+}
+
 // src/rag/embedding.ts
 function cosineSimilarity(left, right) {
   if (left.length === 0 || right.length === 0 || left.length !== right.length) {
@@ -1766,6 +1896,20 @@ async function createHybridRetriever(input2) {
     }
   };
 }
+async function createAgenticRagSearchTools(input2) {
+  const episodes = new EpisodeFtsRetriever(new EpisodeRepository(input2.database));
+  const messages = new MessageFtsRetriever(input2.messages, { excludeMessageIds: input2.excludeMessageIds });
+  const semantic = hasEmbeddingConfig(input2.config, input2.secrets) ? new VectorRetriever(
+    createEmbeddingModel(input2.config, input2.secrets),
+    new SqliteVectorStore(input2.database, { model: input2.config.embedding.model })
+  ) : void 0;
+  const hybrid = new HybridRetriever(semantic ? [episodes, messages, semantic] : [episodes, messages]);
+  return {
+    tools: createRagSearchTools({ hybrid, messages, episodes, semantic }),
+    close: () => {
+    }
+  };
+}
 
 // src/doctor/checks.ts
 function pass(name, message) {
@@ -2454,12 +2598,99 @@ async function generateGroundedAnswer(input2) {
   };
 }
 
-// src/rag/qa-service.ts
-async function askWithRag(input2) {
-  const evidence = await input2.retriever.retrieve(input2.question);
+// src/rag/agentic-qa-service.ts
+var DEFAULT_MAX_MODEL_TURNS = 4;
+var DEFAULT_MAX_TOOL_CALLS = 8;
+var DEFAULT_MAX_EVIDENCE = 12;
+var NO_EVIDENCE_ANSWER = "\u4E0D\u77E5\u9053\u3002\u5F53\u524D\u672C\u5730\u77E5\u8BC6\u5E93\u6CA1\u6709\u68C0\u7D22\u5230\u8DB3\u591F\u8BC1\u636E\u3002";
+var AGENTIC_SYSTEM_PROMPT = "\u4F60\u662F\u672C\u5730\u77E5\u8BC6\u4FE1\u606F\u6536\u96C6\u4EE3\u7406\u3002\u4F60\u7684\u804C\u8D23\u662F\u56F4\u7ED5\u7528\u6237\u95EE\u9898\u51B3\u5B9A\u662F\u5426\u8C03\u7528\u641C\u7D22\u5DE5\u5177\u3001\u9009\u62E9\u5408\u9002\u7684\u5DE5\u5177\u548C\u67E5\u8BE2\u8BCD\uFF0C\u5E76\u6839\u636E\u5F53\u524D\u7ED3\u679C\u51B3\u5B9A\u662F\u5426\u7EE7\u7EED\u641C\u7D22\u3002\u4E0D\u8981\u7F16\u9020\u4EFB\u4F55\u8BC1\u636E\u6216\u58F0\u79F0\u770B\u8FC7\u672A\u68C0\u7D22\u5230\u7684\u5185\u5BB9\u3002\u4F60\u7684\u8F93\u51FA\u53EA\u7528\u4E8E\u6536\u96C6\u8BC1\u636E\uFF0C\u6700\u7EC8\u7B54\u6848\u4F1A\u7531\u53E6\u4E00\u4E2A\u57FA\u4E8E\u8BC1\u636E\u7684\u6B65\u9AA4\u751F\u6210\u3002";
+function toToolResultContent(results) {
+  return JSON.stringify(
+    results.map((item) => ({
+      id: item.id,
+      text: item.text,
+      score: item.score,
+      source: item.source
+    }))
+  );
+}
+function toToolErrorContent(message) {
+  return JSON.stringify({ error: message });
+}
+function dedupeEvidence(evidence, maxEvidence) {
+  const deduped = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const item of evidence) {
+    if (seen.has(item.id)) {
+      continue;
+    }
+    seen.add(item.id);
+    deduped.push(item);
+    if (deduped.length >= maxEvidence) {
+      break;
+    }
+  }
+  return deduped;
+}
+async function askWithAgenticRag(input2) {
+  if (!input2.model.completeWithTools) {
+    throw new Error("\u5F53\u524D LLM \u5BA2\u6237\u7AEF\u4E0D\u652F\u6301\u5DE5\u5177\u8C03\u7528\u3002");
+  }
+  const maxModelTurns = input2.maxModelTurns ?? DEFAULT_MAX_MODEL_TURNS;
+  const maxToolCalls = input2.maxToolCalls ?? DEFAULT_MAX_TOOL_CALLS;
+  const maxEvidence = input2.maxEvidence ?? DEFAULT_MAX_EVIDENCE;
+  const messages = [
+    { role: "system", content: AGENTIC_SYSTEM_PROMPT },
+    { role: "user", content: input2.question }
+  ];
+  const toolsByName = new Map(input2.tools.map((tool) => [tool.name, tool]));
+  let evidence = [];
+  let toolCallsUsed = 0;
+  for (let turn = 0; turn < maxModelTurns; turn += 1) {
+    const assistantResult = await input2.model.completeWithTools(messages, input2.tools);
+    messages.push({
+      role: "assistant",
+      content: assistantResult.content,
+      toolCalls: assistantResult.toolCalls
+    });
+    if (assistantResult.toolCalls.length === 0) {
+      break;
+    }
+    for (const toolCall of assistantResult.toolCalls) {
+      if (toolCallsUsed >= maxToolCalls) {
+        break;
+      }
+      toolCallsUsed += 1;
+      const tool = toolsByName.get(toolCall.name);
+      if (!tool) {
+        messages.push({
+          role: "tool",
+          toolCallId: toolCall.id,
+          content: toToolErrorContent(`\u672A\u77E5\u5DE5\u5177\uFF1A${toolCall.name}`)
+        });
+        continue;
+      }
+      try {
+        const results = await tool.execute(toolCall.input);
+        evidence = dedupeEvidence([...evidence, ...results], maxEvidence);
+        messages.push({
+          role: "tool",
+          toolCallId: toolCall.id,
+          content: toToolResultContent(results)
+        });
+      } catch (error) {
+        const message = error instanceof Error ? error.message : String(error);
+        messages.push({
+          role: "tool",
+          toolCallId: toolCall.id,
+          content: toToolErrorContent(message)
+        });
+      }
+    }
+  }
   if (evidence.length === 0) {
     return {
-      answer: "\u4E0D\u77E5\u9053\u3002\u5F53\u524D\u672C\u5730\u77E5\u8BC6\u5E93\u6CA1\u6709\u68C0\u7D22\u5230\u8DB3\u591F\u8BC1\u636E\u3002",
+      answer: NO_EVIDENCE_ANSWER,
       citations: []
     };
   }
@@ -2577,7 +2808,7 @@ var FeishuQuestionHandler = class {
     }
     const questionMessageId = payload.event?.message?.message_id;
     await this.acknowledgeQuestion(decision.chatId, questionMessageId);
-    const { retriever, close } = await createHybridRetriever({
+    const { tools, close } = await createAgenticRagSearchTools({
       config: this.options.config,
       secrets: this.options.secrets,
       database: this.options.database,
@@ -2586,9 +2817,9 @@ var FeishuQuestionHandler = class {
     });
     try {
       try {
-        const result = await askWithRag({
+        const result = await askWithAgenticRag({
           question: decision.question,
-          retriever,
+          tools,
           model: this.options.model
         });
         const citations = formatCitations(result.citations);
@@ -3433,6 +3664,22 @@ async function processMessagesNow(input2) {
   };
 }
 
+// src/rag/qa-service.ts
+async function askWithRag(input2) {
+  const evidence = await input2.retriever.retrieve(input2.question);
+  if (evidence.length === 0) {
+    return {
+      answer: "\u4E0D\u77E5\u9053\u3002\u5F53\u524D\u672C\u5730\u77E5\u8BC6\u5E93\u6CA1\u6709\u68C0\u7D22\u5230\u8DB3\u591F\u8BC1\u636E\u3002",
+      citations: []
+    };
+  }
+  return generateGroundedAnswer({
+    question: input2.question,
+    evidence,
+    model: input2.model
+  });
+}
+
 // src/update/npm-updater.ts
 import { execFile } from "child_process";
 import { promisify } from "util";