A pi extension that bridges WeChat to your Pi coding agent. Scan a QR code in WeChat → chat with Pi from your phone. Supports text, images, files, voice, and video. Uses the @wechatbot/wechatbot SDK for all iLink protocol operations.

How it works

graph LR
  A["📱 WeChat User<br/>(phone)"] <--> B["☁️ iLink API<br/>(Tencent)"]
  B <--> C["🤖 Pi Agent<br/>(laptop)"]
  B --- D["@wechatbot/wechatbot SDK"]
  style D fill:none,stroke:none,color:#666

You run /wechat in pi
Extension creates a WeChatBot instance → calls bot.login()
SDK calls iLink API → gets a QR code URL → displayed in pi terminal
You scan the QR with WeChat on your phone
WeChat messages → pi.sendUserMessage() → become pi prompts
pi.on('agent_end') → bot.reply() → sent back to WeChat
Typing indicator shown while Pi thinks

Install

From npm (recommended):

1pi install npm:@wechatbot/pi-agent

Or load directly (no install):

1pi -e npm:@wechatbot/pi-agent

Or copy to extensions directory (auto-loads):

1cp -r pi-agent/ ~/.pi/agent/extensions/wechat-bridge/

Usage

1234567891011> /wechat

  📱 Scan this QR code in WeChat:

    ▄▄▄▄▄▄▄ ▄▄▄ ▄▄▄▄▄▄▄
    █ ▄▄▄ █ █▀█ █ ▄▄▄ █
    ...

[wechat] QR scanned — confirm in WeChat
[wechat] Login confirmed
✓ WeChat connected! Account: e06c1ceea05e@im.bot

Now send a message from WeChat — Pi receives it as a prompt, processes it, and sends the reply back.

Message flow

WeChat user sends "帮我重构 auth 模块"
       │
       ▼
iLink API (long-poll getupdates)
       │
       ▼
Extension receives IncomingMessage
       │
       ├── bot.sendTyping(userId)        → "对方正在输入中..."
       │
       ├── pi.sendUserMessage(content)   → becomes pi prompt
       │
       ▼
Pi agent processes (tools, thinking...)
       │
       ▼
pi.on('agent_end') fires
       │
       ├── Extract assistant text
       ├── stripMarkdown()               → clean for WeChat
       ├── Detect generated file paths   → send as media
       ├── bot.stopTyping(userId)
       └── bot.reply(msg, content)       → sent back to WeChat

Media support

The extension handles all WeChat message types — not just text.

Inbound (WeChat → Pi)

Type	How it's handled
Text	Sent directly as pi prompt
Image	Downloaded → base64 → sent as ImageContent for vision models
Voice	Uses WeChat transcription if available; otherwise downloads SILK audio
File	Text files (.txt, .md, .py, etc): content included in prompt. Others: described with metadata
Video	Downloaded → saved to temp file → pi told the file path for processing

Outbound (Pi → WeChat)

Type	How it's handled
Text	Markdown stripped → sent as plain text
Generated files	Detected in response → auto-sent via bot.reply({ file })
Auto-routing	.png → image message, .mp4 → video message, .pdf → file attachment

Commands

/wechat              Start login (show QR code)
/wechat --force      Force re-login (new QR code)
/wechat-disconnect   Disconnect WeChat
/wechat-send <text>  Manually send text to WeChat user

When already connected, /wechat shows a menu: Disconnect, Show status, or Cancel.

Pi APIs used

Extension API usage

Pi API	Purpose	When
pi.registerCommand()	/wechat, /wechat-send, /wechat-disconnect	Extension load
pi.sendUserMessage()	WeChat msg → pi prompt (text or image+text)	On each message
pi.on(agent_end)	Capture response → stripMarkdown → send to WeChat	After agent finishes
pi.on(message_update)	Track streaming text	During streaming
ctx.ui.setStatus()	Connection status in footer	Throughout
pi.on(session_shutdown)	Cleanup on exit	On shutdown

Features

Zero config — just /wechat and scan
Full media support — images (vision), voice, files, video
Markdown stripping — AI output → clean plain text for WeChat
Auto media routing — generated .png sent as image, .mp4 as video, etc.
Credential persistence — ~/.wechatbot/, skips QR on restart
Session recovery — auto re-login on -14
Typing indicator — "对方正在输入中" while Pi thinks
Text chunking — long replies split at 2000 chars
Graceful shutdown — stops on session_shutdown