@bolloon/bolloon-agent 0.1.35 → 0.1.37

package/src/agents/p2p-document-tools.ts

@@ -1,347 +0,0 @@
-/**
- * P2P Document Tools - Tools for sending/receiving documents over iroh P2P
- */
-
-import * as crypto from 'crypto';
-import * as fs from 'fs/promises';
-import * as path from 'path';
-import { irohTransport } from '../network/iroh-transport.js';
-import { documentReader, type DocumentContent } from '../documents/reader.js';
-import { documentStore, type DocumentChunk, type ReceivedDocument } from '../documents/store.js';
-import type { Tool } from './pi-sdk.js';
-
-const CHUNK_SIZE = 60 * 1024;  // 60KB per chunk
-
-function generateDocId(): string {
-  return crypto.randomUUID();
-}
-
-function getMimeType(filename: string): string {
-  const ext = filename.toLowerCase().split('.').pop();
-  const mimeTypes: Record<string, string> = {
-    'txt': 'text/plain',
-    'md': 'text/markdown',
-    'html': 'text/html',
-    'htm': 'text/html',
-    'yaml': 'application/yaml',
-    'yml': 'application/yaml',
-    'json': 'application/json',
-    'pdf': 'application/pdf',
-    'docx': 'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
-  };
-  return mimeTypes[ext || ''] || 'application/octet-stream';
-}
-
-export async function initDocumentReceiver(): Promise<void> {
-  await documentStore.initialize();
-
-  documentStore.onDocumentReceived(async (doc) => {
-    console.log(`[DocumentReceiver] Document received: ${doc.fileName} from ${doc.fromNodeIdShort}`);
-    // 异步调用 LLM 解析（不阻塞接收流程）
-    void parseDocumentWithLLM(doc);
-  });
-
-  irohTransport.onMessage('document_chunk', async (msg) => {
-    try {
-      const chunk: DocumentChunk = JSON.parse(new TextDecoder().decode(msg.payload));
-      const result = await documentStore.receiveChunk(chunk);
-      if (result) {
-        console.log(`[DocumentReceiver] Document complete: ${result.fileName}`);
-      }
-    } catch (e) {
-      console.error('[DocumentReceiver] Failed to process chunk:', e);
-    }
-  });
-
-  console.log('[DocumentReceiver] Initialized and listening for document chunks');
-}
-
-// ============================================================================
-// AI 解析反馈：接收方解析完文档后把摘要回送给发送方
-// ============================================================================
-
-export interface AIFeedbackMessage {
-  docId: string;
-  fileName: string;
-  mimeType: string;
-  summary: string;
-  qualityScore: number;
-  feedbackAt: number;
-  fromNodeId: string;
-}
-
-export type AIFeedbackHandler = (feedback: AIFeedbackMessage, fromNodeId: string) => void;
-const feedbackHandlers: Set<AIFeedbackHandler> = new Set();
-
-/** 注册 AI 解析反馈监听器（发送方调用） */
-export function onAIFeedback(handler: AIFeedbackHandler): void {
-  feedbackHandlers.add(handler);
-  // iroh listener 只挂一次
-  ensureFeedbackListenerInstalled();
-}
-
-let feedbackListenerInstalled = false;
-function ensureFeedbackListenerInstalled(): void {
-  if (feedbackListenerInstalled) return;
-  feedbackListenerInstalled = true;
-  irohTransport.onMessage('ai_feedback', (msg) => {
-    try {
-      const fb: AIFeedbackMessage = JSON.parse(new TextDecoder().decode(msg.payload));
-      for (const h of feedbackHandlers) {
-        try {
-          h(fb, msg.from);
-        } catch (e) {
-          console.error('[AIFeedback] handler error:', e);
-        }
-      }
-    } catch (e) {
-      console.error('[AIFeedback] failed to parse message:', e);
-    }
-  });
-}
-
-/** 文档解析服务：优先走 web 统一入口 (含 judgment + harness), fallback 到本地 LLM */
-async function callAIParseService(text: string, fileName: string, mimeType: string, fromNodeId?: string): Promise<{ summary: string; qualityScore: number; source: 'web' | 'local'; judgmentId?: string; gateArtifact?: string }> {
-  // 1) 优先: 调 web 端 POST /api/ai-parse (统一入口, 含 judgment + harness)
-  const webBase = process.env.BOLLOON_WEB_URL || process.env.PORTAL_URL || 'http://127.0.0.1:54188';
-  try {
-    const r = await fetch(`${webBase}/api/ai-parse`, {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ text, mimeType, fileName, fromNodeId, source: 'p2p-document' }),
-      // 短超时, fallback 友好
-      signal: AbortSignal.timeout(8000),
-    });
-    if (r.ok) {
-      const data = await r.json() as any;
-      return {
-        summary: data.summary,
-        qualityScore: data.qualityScore,
-        source: 'web',
-        judgmentId: data.judgmentId,
-        gateArtifact: data.gateArtifact,
-      };
-    }
-    console.warn(`[AIParse] web returned ${r.status}, fallback to local`);
-  } catch (e) {
-    console.warn(`[AIParse] web ${webBase} unreachable (${(e as Error).message}), fallback to local`);
-  }
-
-  // 2) Fallback: 直接调本地 LLM (不调 judgment/harness, 仅在 web 不可用时使用)
-  const apiKey = process.env.OPENAI_API_KEY || process.env.ANTHROPIC_API_KEY || process.env.MINIMAX_API_KEY || process.env.OPENROUTER_API_KEY || process.env.GEMINI_API_KEY;
-  if (!apiKey) {
-    throw new Error('No web endpoint and no LLM API key configured. Set BOLLOON_WEB_URL or OPENAI_API_KEY.');
-  }
-  const { getMinimax } = await import('../constraints/index.js');
-  const llm = getMinimax();
-  const truncated = text.length > 6000 ? text.substring(0, 6000) + '...[截断]' : text;
-  const prompt = `请分析以下 ${mimeType} 文档，并给出 (1) 一句话中文摘要 (2) 关键要点列表 (3) 文档质量评分(0-1)。\n\n文件名: ${fileName}\n\n内容:\n${truncated}`;
-  const r = await llm.summarize(prompt);
-  return { summary: r.summary, qualityScore: r.qualityScore, source: 'local' };
-}
-
-async function parseDocumentWithLLM(doc: ReceivedDocument): Promise<void> {
-  try {
-    const filePath = path.join(
-      process.env.HOME || '/tmp',
-      '.bolloon', 'documents', 'received',
-      doc.id, doc.fileName
-    );
-    const content = await fs.readFile(filePath, 'utf-8');
-
-    const r = await callAIParseService(content, doc.fileName, doc.mimeType, doc.fromNodeId);
-
-    const sidecar = {
-      filename: doc.fileName,
-      mimeType: doc.mimeType,
-      fromNodeId: doc.fromNodeId,
-      receivedAt: doc.receivedAt,
-      summary: r.summary,
-      qualityScore: r.qualityScore,
-      parseSource: r.source,
-      judgmentId: r.judgmentId,
-      gateArtifact: r.gateArtifact,
-      analyzedAt: Date.now(),
-    };
-    const sidecarPath = path.join(
-      process.env.HOME || '/tmp',
-      '.bolloon', 'documents', 'received',
-      doc.id, 'ai-analysis.json'
-    );
-    await fs.writeFile(sidecarPath, JSON.stringify(sidecar, null, 2));
-    console.log(`[DocumentReceiver] AI parsed ${doc.fileName} via ${r.source} (score=${r.qualityScore.toFixed(2)})${r.judgmentId ? ` judgment=${r.judgmentId}` : ''}${r.gateArtifact ? ` gate=${r.gateArtifact}` : ''}`);
-
-    // 把 AI 解析结果回送给发送方 (doc.fromNodeId)，形成 "接收 → 解析 → 反馈" 闭环
-    if (doc.fromNodeId) {
-      const feedback: AIFeedbackMessage = {
-        docId: doc.id,
-        fileName: doc.fileName,
-        mimeType: doc.mimeType,
-        summary: r.summary,
-        qualityScore: r.qualityScore,
-        feedbackAt: Date.now(),
-        fromNodeId: irohTransport.getNodeId() || '',
-      };
-      const sent = await irohTransport.sendMessage(
-        doc.fromNodeId,
-        'ai_feedback',
-        new TextEncoder().encode(JSON.stringify(feedback))
-      );
-      if (sent) {
-        console.log(`[DocumentReceiver] Feedback sent to ${doc.fromNodeIdShort}`);
-      } else {
-        console.warn(`[DocumentReceiver] Failed to send feedback to ${doc.fromNodeIdShort}`);
-      }
-    }
-  } catch (e) {
-    console.error(`[DocumentReceiver] AI parse failed for ${doc.fileName}:`, e);
-  }
-}
-
-export const p2pDocumentTools: Tool[] = [
-  {
-    name: 'list_online_peers',
-    description: '列出当前通过 iroh P2P 网络在线的对等节点',
-    parameters: {},
-    execute: async () => {
-      try {
-        const peers = irohTransport.getConnectedPeers();
-        if (peers.length === 0) {
-          return { success: true, output: '当前无在线的对等节点' };
-        }
-        return {
-          success: true,
-          output: `在线节点 (${peers.length}):\n${peers.map(p => `  - ${p.substring(0, 16)}...`).join('\n')}`
-        };
-      } catch (e) {
-        return { success: false, error: String(e) };
-      }
-    }
-  },
-
-  {
-    name: 'send_document',
-    description: '读取本地文档并发送给指定的对等节点，支持 .txt, .md, .pdf, .docx 格式',
-    parameters: {
-      target_peer_id: '目标对等节点的完整 nodeId',
-      file_path: '要发送的本地文件路径',
-      message: '可选的附言消息'
-    },
-    execute: async (args) => {
-      try {
-        const { target_peer_id, file_path, message } = args;
-
-        // 读取文档
-        let content: DocumentContent;
-        try {
-          content = await documentReader.read(file_path);
-        } catch (e) {
-          return { success: false, error: `无法读取文件: ${e}` };
-        }
-
-        // 分块
-        const mimeType = getMimeType(file_path);
-        const fileData = await fs.readFile(file_path);
-        const contentBase64 = fileData.toString('base64');
-        const totalChunks = Math.ceil(contentBase64.length / CHUNK_SIZE);
-
-        const docId = generateDocId();
-
-        // 逐块发送
-        for (let i = 0; i < totalChunks; i++) {
-          const chunk: DocumentChunk = {
-            docId,
-            fileName: content.metadata.filename,
-            fileSize: content.metadata.size,
-            mimeType,
-            chunkIndex: i,
-            totalChunks,
-            content: contentBase64.substring(i * CHUNK_SIZE, (i + 1) * CHUNK_SIZE),
-            fromNodeId: irohTransport.getNodeId() || '',
-            message: i === 0 ? message : undefined,
-            timestamp: Date.now(),
-          };
-
-          const sent = await irohTransport.sendMessage(
-            target_peer_id,
-            'document_chunk',
-            new TextEncoder().encode(JSON.stringify(chunk))
-          );
-
-          if (!sent) {
-            return { success: false, error: `发送第 ${i + 1}/${totalChunks} 块失败` };
-          }
-        }
-
-        return {
-          success: true,
-          output: `📤 文档已发送: ${content.metadata.filename}\n大小: ${content.metadata.size} 字节\n分块: ${totalChunks}\n目标: ${target_peer_id.substring(0, 16)}...`
-        };
-      } catch (e) {
-        return { success: false, error: String(e) };
-      }
-    }
-  },
-
-  {
-    name: 'receive_documents',
-    description: '列出已接收到的文档，可按发送者筛选',
-    parameters: {
-      limit: '返回数量上限，默认 50',
-      sender_peer_id: '可选，按发送者 nodeId 筛选'
-    },
-    execute: async (args) => {
-      try {
-        const limit = parseInt(args.limit) || 50;
-        const docs = await documentStore.getReceivedDocuments(limit, args.sender_peer_id);
-
-        if (docs.length === 0) {
-          return { success: true, output: '暂无已接收的文档' };
-        }
-
-        const list = docs.map(d =>
-          `📄 ${d.fileName}\n   ID: ${d.id}\n   大小: ${d.fileSize} 字节\n   来自: ${d.fromNodeIdShort}\n   时间: ${new Date(d.receivedAt).toLocaleString()}`
-        ).join('\n\n');
-
-        return {
-          success: true,
-          output: `已接收文档 (${docs.length}):\n\n${list}`
-        };
-      } catch (e) {
-        return { success: false, error: String(e) };
-      }
-    }
-  },
-
-  {
-    name: 'read_received_document',
-    description: '读取已接收文档的内容，返回文档ID可查询文档详情',
-    parameters: {
-      document_id: '文档ID (从 receive_documents 获取)'
-    },
-    execute: async (args) => {
-      try {
-        const { document_id } = args;
-
-        if (!document_id) {
-          return { success: false, error: '缺少 document_id 参数' };
-        }
-
-        const result = await documentStore.readDocument(document_id);
-
-        if (!result) {
-          return { success: false, error: `未找到文档: ${document_id}` };
-        }
-
-        const { content, metadata } = result;
-
-        return {
-          success: true,
-          output: `📄 ${metadata.fileName}\nID: ${metadata.id}\n大小: ${metadata.fileSize} 字节\n来自: ${metadata.fromNodeIdShort}\n\n${content.substring(0, 2000)}${content.length > 2000 ? '...\n(内容已截断)' : ''}`
-        };
-      } catch (e) {
-        return { success: false, error: String(e) };
-      }
-    }
-  }
-];