什么是 io.github.elagerway/magpipe?
通过 Magpipe 管理 AI 语音代理、通话、SMS、联系人和电话号码。
README
Magpipe - A Conversational Platform
Magpipe is a a conversational platform that manages SMS/Text messsges, Phone Calls, Email, and Web Chat. Built with vanilla JavaScript, Supabase, CPaaS, and LiveKit.
Features
- Smart Call Handling — AI answers calls, screens unknown callers, takes messages, and transfers important calls to you
- Intelligent SMS — Context-aware text responses that understand conversation history and respond appropriately
- Email AI — Automatically respond to emails with context-aware AI replies, drafts for review, or fully autonomous mode
- 24/7 Availability — Your assistant never sleeps. Handle calls and messages around the clock
- Contact Management — Whitelist VIPs, block spam, and set custom rules for different callers
- Knowledge Base — Train your assistant with custom information to answer questions accurately
- Privacy First — Your data is encrypted and secure. Full control over what information is shared
- Analytics & Insights — Real-time dashboards with call volume, sentiment analysis, and conversation trends
- Real-Time Translation — Automatic translation across 30+ languages during live calls
- Conversation Memory — Your agent remembers past interactions, caller preferences, and context across every conversation
- Integrations — Connect with HubSpot, Google Workspace, Zapier, and more. Sync contacts, log calls, and automate workflows
- Warm Transfer — Seamlessly hand off live calls to you or your team when the AI detects a conversation needs a human touch
- Social Listening — Automated monitoring of Reddit, HackerNews, and Google for keyword mentions. Email digests, admin dashboard with favorites, search, and status tracking
- Voice Library — 30 voice options (24 ElevenLabs + 6 OpenAI) with AI-generated avatars, audio previews, trait tags (accent, tone, age, use case), and multilingual support. Cloned voices use the Multilingual v2 model for natural accent preservation
- Blog & Social — Built-in blog with AI-generated featured images, Twitter/X auto-posting with image attachments, and a What's New notification card for users
Tech Stack
- Frontend: Vanilla JavaScript (ES6+), HTML5, CSS3, Vite
- Backend: Supabase (PostgreSQL, Auth, Realtime, Edge Functions — TypeScript/Deno)
- Telephony: SignalWire (PSTN calls, SMS, conference bridging)
- Voice AI: LiveKit (real-time audio), OpenAI (LLM), Deepgram (STT), ElevenLabs (TTS)
- Payments: Stripe (credits, subscriptions)
- Email: Postmark (transactional), Gmail API (support tickets, inbox email send/receive)
- Enrichment: Apollo.io (contact data)
- Social Listening: Serper.dev (Google SERP + Reddit), HN Algolia API
- Vector DB: pgvector (conversation context embeddings)
Project Structure
pat/
├── agents/
│ └── livekit-voice-agent/ # Python LiveKit agent (deployed on Render)
│ ├── agent.py # Main agent logic (~3,400 lines)
│ ├── requirements.txt
│ └── render.yaml # Render deployment config
├── public/ # Static assets
│ ├── index.html
│ ├── manifest.json # PWA manifest
│ ├── sw.js # Service worker
│ └── styles/
├── src/
│ ├── components/ # Reusable UI components
│ │ ├── AdminChatInterface.js # Admin AI chat
│ │ ├── AdminHeader.js # Admin portal header with status dropdown
│ │ ├── BottomNav.js # Mobile bottom nav / desktop sidebar
│ │ ├── ConfirmModal.js # Custom confirm dialog
│ │ ├── ImpersonationBanner.js
│ │ ├── LowBalanceBanner.js # Low credit balance warning
│ │ └── ...
│ ├── lib/ # Utilities
│ │ └── supabase.js # Supabase client
│ ├── models/ # Data models (User, Contact, AgentConfig, etc.)
│ ├── pages/ # UI pages (large pages split into subdirectories)
│ │ ├── admin/ # Admin portal (split from admin.js)
│ │ │ ├── index.js # Entry point
│ │ │ ├── support-tab.js # Support tickets, Gmail, AI settings
│ │ │ ├── users-tab.js # User management
│ │ │ ├── analytics-tab.js
│ │ │ ├── kpi-tab.js
│ │ │ ├── notifications-tab.js
│ │ │ ├── chat-tab.js
│ │ │ ├── global-agent-tab.js
│ │ │ └── styles.js
│ │ ├── agent-detail/ # Agent config (split from agent-detail.js)
│ │ │ ├── index.js
│ │ │ ├── configure-tab.js
│ │ │ ├── prompt-tab.js
│ │ │ ├── functions-tab.js
│ │ │ ├── knowledge-tab.js
│ │ │ ├── memory-tab.js
│ │ │ ├── deployment-tab.js # Phone number assignment + inline buy-number modal
│ │ │ └── ...
│ │ ├── inbox/ # Inbox (split from inbox.js)
│ │ │ ├── index.js
│ │ │ ├── call-interface.js
│ │ │ ├── listeners.js
│ │ │ ├── messaging.js
│ │ │ └── views.js
│ │ ├── phone/ # Phone (split from phone.js)
│ │ │ ├── index.js
│ │ │ ├── call-handler.js
│ │ │ ├── dialpad.js
│ │ │ └── number-management.js
│ │ ├── batch-calls.js # Batch outbound calls (CSV upload, scheduling)
│ │ ├── contacts.js
│ │ ├── settings.js
│ │ └── ...
│ ├── services/ # Business logic services
│ │ ├── unreadService.js
│ │ ├── pushNotifications.js
│ │ ├── mcpClient.js
│ │ └── ...
│ ├── main.js # App entry point
│ └── router.js # Client-side SPA routing
├── supabase/
│ ├── functions/ # ~160 Edge Functions (TypeScript/Deno)
│ │ ├── _shared/ # Shared utilities (cors, auth, balance-check)
│ │ ├── admin-status/
│ │ ├── contact-lookup/ # Apollo.io enrichment
│ │ ├── batch-calls/ # Batch call CRUD (create, list, get, start, cancel)
│ │ ├── process-batch-calls/ # Batch call worker (initiates calls per recipient)
│ │ ├── initiate-bridged-call/
│ │ ├── gmail-push-webhook/ # Gmail Pub/Sub push notifications + support ticket mirroring
│ │ ├── gmail-watch-renew/ # Daily Gmail watch renewal
│ │ ├── poll-gmail-inbox/ # Fallback email polling (30m)
│ │ ├── poll-gmail-tickets/
│ │ ├── support-tickets-api/
│ │ ├── webhook-inbound-call/
│ │ ├── webhook-inbound-sms/
│ │ └── ...
│ └── migrations/ # Database migrations
├── packages/
│ └── mcp-server/ # MCP server for AI coding tools (Claude Code, Cursor)
├── tests/ # Playwright & contract tests
└── specs/ # Feature specifications
Getting Started
Prerequisites
- Node.js 18+ and npm
- Supabase account & CLI
- SignalWire account (SIP endpoint, phone numbers)
- LiveKit Cloud account
- OpenAI API key
- Deepgram API key
- ElevenLabs API key
Installation
-
Clone the repository
bashgit clone <repo-url> cd pat -
Install dependencies
bashnpm install -
Set up environment variables
bashcp .env.example .envEdit
.envwith your credentials:code# Supabase VITE_SUPABASE_URL=your_supabase_url VITE_SUPABASE_ANON_KEY=your_supabase_anon_key SUPABASE_SERVICE_ROLE_KEY=your_service_role_key # SignalWire SIGNALWIRE_PROJECT_ID=your_project_id SIGNALWIRE_API_TOKEN=your_api_token SIGNALWIRE_SPACE_URL=your_space.signalwire.com # LiveKit LIVEKIT_URL=wss://your-project.livekit.cloud LIVEKIT_API_KEY=your_api_key LIVEKIT_API_SECRET=your_api_secret # AI Services OPENAI_API_KEY=your_openai_key DEEPGRAM_API_KEY=your_deepgram_key ELEVENLABS_API_KEY=your_elevenlabs_key -
Run database migrations
bashnpx supabase db push -
Deploy Edge Functions
bashexport SUPABASE_ACCESS_TOKEN=your_token npx supabase functions deploy --no-verify-jwt -
Start development server
bashnpm run devThe app will be available at
http://localhost:3000
Architecture
Smart Call Handling
Inbound calls hit a SignalWire number, which triggers a LiveKit room. The AI agent joins, screens the caller, and decides whether to transfer, take a message, or handle the conversation. All calls are recorded with transcripts stored automatically.
Intelligent SMS
Inbound SMS triggers a webhook that retrieves conversation context (with vector embeddings), generates a context-aware AI response, and sends it via SignalWire. Full conversation history is maintained per contact.
Only agent_type = 'text' agents handle SMS. Voice agents are never routed inbound SMS. Agent selection priority: explicit text_agent_id on the service number → user's default text agent → any text agent belonging to the user. If no text agent is found, a one-time auto-reply is sent to the sender and subsequent messages from that number are silently ignored.
Loop detection prevents runaway back-and-forth: if the same message body arrives more than twice in a conversation thread, the AI reply is skipped silently. Normal replies resume as soon as the sender sends a different message.
Email AI
Gmail integration via OAuth and Pub/Sub push notifications. The agent can operate in draft mode (generates replies for review) or auto mode (sends replies autonomously). Emails are threaded and visible in the unified inbox.
Warm Transfer
During a live call, the AI can seamlessly hand off to a human. SignalWire bridges the caller to your phone while the agent briefs you, then drops off the line.
Knowledge Base
Upload documents or URLs to train your agent. Content is crawled (via Firecrawl), chunked, and embedded with OpenAI into pgvector for RAG-powered retrieval during calls and messages.
Conversation Memory
Every interaction is summarized and stored with vector embeddings. The agent retrieves relevant past context at the start of each new conversation for continuity across calls, SMS, and email.
Real-Time Translation
The voice agent supports automatic translation across 30+ languages during live calls, allowing callers and agents to speak in different languages.
Analytics & Insights
Real-time dashboards display call volume, sentiment analysis, and conversation trends. Organization-wide analytics are available for teams.
Integrations
HubSpot (CRM), Google Workspace (Calendar, Gmail), Cal.com (booking), Apollo.io (contact enrichment), and Slack (notifications). Native tool execution via the agent chat interface.
Contact Management
Contacts can be whitelisted for priority handling, blocked, or given custom call rules. CSV import, Apollo.io enrichment, and automatic contact creation from calls/SMS.
Privacy & Security
All data is encrypted in transit and at rest via Supabase. Access codes, phone verification, and role-based permissions control who can access what.
24/7 Availability
The LiveKit voice agent runs on Render and connects to LiveKit Cloud, handling calls and messages around the clock without downtime.
Database Schema
Core Tables
users- User profiles, SIP credentials, preferences, credit balanceorganizations- Multi-user org managementcontacts- Contact information with enrichment fieldsagent_configs- AI agent configuration (5 types: inbound_voice, outbound_voice, text, email, chat_widget; single system_prompt with type-specific defaults; shared_memory_agent_ids for cross-agent memory sharing; voice settings, functions)call_records- Call history with purpose, goal, transcriptssms_messages- SMS message historyemail_messages- Email threads for inbox (id, user_id, agent_id, contact_id, thread_id, gmail_message_id, from/to/cc/bcc, subject, body_text, body_html, direction, status, is_ai_generated, is_read, sent_at,attachmentsJSONB for image metadata)agent_email_configs- Per-agent email channel config (gmail_address, agent_mode: off/draft/auto, watch_expiration, last_history_id)conversation_contexts- Conversation memory with embeddingsservice_numbers- User's phone numbers from SignalWire (agent_idfor voice,text_agent_idfor SMS — independent routing)support_tickets- Support ticket threads (Gmail integration, contact form feedback,ticket_refTKT-XXXXXX,attachmentsJSONB for image metadata)support_ticket_notes- Internal notes on support threadssupport_email_config- Gmail connection, AI agent settings, ticket creation togglesupport-attachments(storage bucket) - Image attachments uploaded from Gmail or admin UIcredit_transactions- Billing and credit historyknowledge_sources/knowledge_chunks- RAG knowledge base (pgvector)chat_widgets/chat_sessions/chat_messages- Embeddable chat widgetbatch_calls- Batch outbound call jobs (status, scheduling, concurrency settings)batch_call_recipients- Per-recipient tracking for batch calls (status, call_record_id)referral_rewards- Referral tracking and bonus payoutssocial_listening_keywords- Tracked keywords for social monitoringsocial_listening_results- Reddit/HackerNews/Google mention results
Deployment
Frontend (Vite)
npm run build
# Deploy dist/ folder to Netlify, Vercel, or similar
Edge Functions (Supabase)
export SUPABASE_ACCESS_TOKEN=your_token
npx supabase functions deploy --no-verify-jwt
LiveKit Agent (Render)
The agent auto-deploys from the master branch:
git push origin master
Development
Available Scripts
npm run dev- Start development servernpm run build- Build for productionnpm run preview- Preview production buildnpm test- Run Playwright testsnpm run lint- Lint code
Testing
# Run all Playwright tests
npx playwright test
# Run specific test
npx playwright test tests/test-outbound-call.spec.js
# Run with UI
npx playwright test --ui
License
MIT
Support
For issues and questions, please open a GitHub issue.
常见问题
io.github.elagerway/magpipe 是什么?
通过 Magpipe 管理 AI 语音代理、通话、SMS、联系人和电话号码。
相关 Skills
网页构建器
by anthropics
面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。
✎ 在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。
网页应用测试
by anthropics
用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。
✎ 借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。
前端设计
by anthropics
面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。
✎ 想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。
相关 MCP Server
GitHub
编辑精选by GitHub
GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。
✎ 这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。
Context7 文档查询
编辑精选by Context7
Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。
✎ 它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。
by tldraw
tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。
✎ 这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。