WhatsApp

openclaw apps install whatsapp

WhatsApp is the world’s most-used messenger, and openclawOS makes it your AI agent’s front door. Pair a phone number, route messages through Pi, get replies on the same device you already carry. Nothing leaves your network except the LLM call.

Why run an AI agent on WhatsApp?

• You’re already there. Two billion people open WhatsApp every day. Your AI assistant should meet you where you are.
• Family-friendly. Show your partner a recipe-suggesting Pi without making them learn a new app.
• Reliable transport. WhatsApp’s notification stack is best-in-class — your agent’s reply arrives even on a flaky network.
• Media-rich. Photos, voice, location, documents all roundtrip without lossy conversions.

How the pairing works

1. Run openclaw apps install whatsapp (or use the Control UI’s Channels tab).
2. A QR code appears in your terminal.
3. Open WhatsApp on your phone → Settings → Linked Devices → Link a Device. Scan it.
4. Done. The Gateway now sees every message the paired number receives.

The pairing persists across restarts. Re-pair only if you logout/clear linked devices on your phone.

Bindings — who gets what

By default the binding routes 1:1 chats to Pi and ignores groups (to avoid noisy mentions). Edit the binding to:

• Listen in specific groups
• Trigger only when @-mentioned
• Run a different agent for different senders
• Auto-translate before processing

The binding language is YAML, lives under ~/.openclaw/bindings/whatsapp.yaml, and hot-reloads.

Privacy posture

WhatsApp messages are E2E-encrypted between phones. When openclawOS pairs as a linked device, it sees decrypted content the same way the WhatsApp web app would. The decrypted bytes stay on your Gateway. Pi’s outbound LLM call ships only what Pi needs to respond — typically just the latest turn or two, not your entire history.

Limits and quirks

• Voice notes: transcribed via Whisper (local or remote). You can disable transcription in config.
• Reactions: Pi can emit emoji reactions as an alternative to text replies (great for “acked”).
• View-once media: respected; openclawOS does not save view-once content.
• Group admin: needed for some channel-management commands but not for read/reply.

Cost

The library is free; you pay for LLM tokens. A modest WhatsApp Pi (Claude Sonnet, light usage) runs about $3–$8 per month in API fees for a single user. Heavy chatters can hit $20–30; vector memory keeps long histories from inflating token usage.