https://github.com/leocodeio/whatsapp-agent
https://github.com/leocodeio/whatsapp-agent
Last synced: 5 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/leocodeio/whatsapp-agent
- Owner: leocodeio
- Created: 2026-02-09T03:30:51.000Z (5 months ago)
- Default Branch: master
- Last Pushed: 2026-02-15T07:10:02.000Z (5 months ago)
- Last Synced: 2026-02-15T13:32:35.737Z (5 months ago)
- Language: TypeScript
- Size: 52.7 KB
- Stars: 0
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Agents: AGENTS.md
Awesome Lists containing this project
README
# WhatsApp Agent
A standalone WhatsApp AI agent built with **Bun.js**, designed to run inside **OpenClaw** and leverage OpenCode's native model routing.
## Setup
### Prerequisites
- [Bun](https://bun.sh) installed
- [OpenCode](https://opencode.ai) installed and configured
- OpenRouter API key (free tier works)
### Install
```bash
cd whatsapp-agent
bun install
```
### Configure
Copy `.env.example` to `.env` and set your provider:
```bash
# Choose your provider
INFERENCE_PROVIDER=openrouter # or 'anthropic'
# Set the API key for your chosen provider
OPENROUTER_API_KEY=sk-or-... # if using openrouter
ANTHROPIC_API_KEY=sk-ant-... # if using anthropic
```
### Run
```bash
# Production
bun run start
# Development (auto-reload)
bun run dev
```
On first run, scan the QR code in terminal with WhatsApp to authenticate.
## Architecture
```
src/
├── core/agent/agents.ts # AI agent — provider-agnostic, manages conversation history
├── providers/opencode/
│ ├── client.ts # OpenCode CLI invocation wrapper
│ └── resolver.ts # Model resolver with automatic fallback
├── integrations/whatsapp/
│ ├── connection.ts # Baileys v7 connection, QR auth, reconnection
│ └── events.ts # Message event bridge
├── config/
│ ├── env.ts # Environment validation
│ └── settings.ts # App constants
├── utils/
│ ├── logger.ts # Structured logging
│ └── errors.ts # Error handling utilities
└── main.ts # Entry point
```
## Inference Providers
### Anthropic (direct)
Set `INFERENCE_PROVIDER=anthropic` with your `ANTHROPIC_API_KEY`.
- Default model: `claude-sonnet-4-20250514`
- Uses Anthropic SDK directly — fastest, most reliable
### OpenRouter (via OpenCode)
Set `INFERENCE_PROVIDER=openrouter` with your `OPENROUTER_API_KEY`.
- Default model: `openrouter/moonshotai/kimi-k2.5`
- Fallback: `openrouter/arcee-ai/trinity-large-preview:free` (free)
- Auto-resolves and falls back if preferred model unavailable
## How It Works
1. WhatsApp message received via Baileys
2. Event bridge extracts text and routes to agent
3. Agent maintains conversation history per chat
4. Agent invokes OpenCode CLI with the resolved model
5. Response sent back to WhatsApp
## OpenClaw Compatibility
This project is designed to run inside OpenClaw's environment:
- Uses `opencode run -m ` for model invocation (no custom HTTP)
- Respects OpenClaw's provider abstraction
- Environment variables managed via OpenClaw's config
## Error Handling
- Retry with exponential backoff (3 attempts)
- Graceful shutdown on SIGINT/SIGTERM
- Structured logging at all levels
- Conversation history limited to 20 messages per chat
## License
MIT