openclaw/README.md

# 📡 Warelay — WhatsApp Relay CLI (Twilio)

Small TypeScript CLI to send, monitor, and webhook WhatsApp messages via Twilio. Supports Tailscale Funnel and config-driven auto-replies.

## Setup

1. `pnpm install`
2. Copy `.env.example` to `.env` and fill in `TWILIO_ACCOUNT_SID`, `TWILIO_AUTH_TOKEN`, and `TWILIO_WHATSAPP_FROM` (use your approved WhatsApp-enabled Twilio number, prefixed with `whatsapp:`).
   - Alternatively, use API keys: `TWILIO_API_KEY` + `TWILIO_API_SECRET` instead of `TWILIO_AUTH_TOKEN`.
   - Optional: `TWILIO_SENDER_SID` to skip auto-discovery of the WhatsApp sender in Twilio.
3. (Optional) Build: `pnpm build` (scripts run directly via tsx, no build required for local use)

## Commands

- Send: `pnpm warelay send --to +15551234567 --message "Hello" --wait 20 --poll 2`
  - `--wait` seconds (default 20) waits for a terminal delivery status; exits non-zero on failed/undelivered/canceled.
  - `--poll` seconds (default 2) sets the polling interval while waiting.
- Monitor (polling): `pnpm warelay monitor` (defaults: 5s interval, 5m lookback)
  - Options: `--interval <seconds>`, `--lookback <minutes>`
- Webhook (push, works well with Tailscale): `pnpm warelay webhook --port 42873 --reply "Got it!"`
  - Points Twilio’s “Incoming Message” webhook to `http://<your-host>:42873/webhook/whatsapp`
  - With Tailscale, expose it: `tailscale serve tcp 42873 127.0.0.1:42873` and use your tailnet IP.
  - Customize path if desired: `--path /hooks/wa`
  - If no `--reply`, auto-reply can be configured via `~/.warelay/warelay.json` (JSON5)
- Webhook/funnel “up”: `pnpm warelay up --port 42873 --path /webhook/whatsapp`
  - Validates Twilio env, confirms `tailscale` binary, enables Tailscale Funnel, starts the webhook, and sets the Twilio incoming webhook to your Funnel URL via the Twilio API (Channels/Senders → fallback to phone number → fallback to messaging service).
  - Requires Tailscale Funnel to be enabled for your tailnet/device (admin setting). If it isn’t enabled, the command will exit with instructions; alternatively expose the webhook via your own tunnel and set the Twilio URL manually.
- Polling mode (no webhooks/funnel): `pnpm warelay poll --interval 5 --lookback 10 --verbose`
  - Useful fallback if Twilio webhook can’t reach you.
  - Still runs config-driven auto-replies (including command-mode/Claude) for new inbound messages.
- Status: `pnpm warelay status --limit 20 --lookback 240`
  - Lists recent sent/received WhatsApp messages (merged and sorted), defaulting to 20 messages from the past 4 hours. Add `--json` for machine-readable output.

## Config-driven auto-replies

Put a JSON5 config at `~/.warelay/warelay.json`. Examples:

```json5
{
  inbound: {
    // Static text reply with templating
    reply: { mode: 'text', text: 'Echo: {{Body}}' }
  }
}

// Command-based reply (stdout becomes the reply)
{
  inbound: {
    reply: {
      mode: 'command',
      command: ['bash', '-lc', 'echo "You said: {{Body}} from {{From}}"']
    }
  }
}
```

### Options reference (JSON5)

- `inbound.allowFrom?: string[]` — optional allowlist of E.164 numbers (no `whatsapp:` prefix). If set, only these senders trigger auto-replies.
- `inbound.reply.mode: "text" | "command"`
  - `text` — send `inbound.reply.text` after templating.
  - `command` — run `inbound.reply.command` (argv array) after templating; trimmed stdout becomes the reply.
- `inbound.reply.text?: string` — used when `mode` is `text`; supports `{{Body}}`, `{{From}}`, `{{To}}`, `{{MessageSid}}`.
- `inbound.reply.command?: string[]` — argv for the command to run; templated per element.
- `inbound.reply.template?: string` — optional string prepended as the second argv element (handy for adding a prompt prefix).
- `inbound.reply.bodyPrefix?: string` — optional string prepended to `Body` before templating (useful to add system instructions, e.g., `You are a helpful assistant running on the user's Mac. User writes messages via WhatsApp and you respond. You want to be concise in your responses, at most 1000 characters.\n\n`).

Example with an allowlist and Claude CLI one-shot (uses a sample number):

```json5
{
  inbound: {
    allowFrom: ["+15551230000"],
    reply: {
      mode: "command",
      command: [
        "claude",
        "--print",
        "--output-format",
        "text",
        "--dangerously-skip-permissions",
        "--system-prompt",
        "You are an auto-reply bot on WhatsApp. Respond concisely.",
        "{{Body}}"
      ]
    }
  }
}
```

During dev you can run without building: `pnpm dev -- <subcommand>` (e.g. `pnpm dev -- send --to +1...`). Auto-replies apply in webhook and polling modes.

## Notes

- Monitor uses polling; webhook mode is push (recommended).
- Stop monitor/webhook with `Ctrl+C`.
- When an auto-reply is triggered (text or command mode), warelay immediately posts a WhatsApp typing indicator tied to the inbound `MessageSid` so the user sees “typing…” while your handler runs.