rmyndharis/OpenWAPublic

Open Source WhatsApp API Gateway

Features • Quick Start • Docs • API • Contributing

OpenWA is a free, open-source WhatsApp API Gateway designed for developers who need full control over their messaging infrastructure—without vendor lock-in or hidden paywalls.

Built on a pluggable architecture, OpenWA lets you select database engines (SQLite/PostgreSQL), backup/migration storage backends (Local/S3), and cache layers (disabled/Redis) through configuration rather than application-code changes. Message media itself is returned inline to API and webhook consumers; it is not automatically persisted to the storage backend.

When you create or edit an operator or viewer API key in the dashboard, you can tick the WhatsApp sessions that key may use.

• No sessions selected — the key can access every session, including ones created later.
• One or more sessions selected — the key can only list, read, and (for operator) manage those sessions. A request naming any other session returns 401; session-filtered lists (sessions, audit, webhook delivery failures) return that key's rows rather than an error; and the key-management routes and the queue dashboard, which name no session at all, return 403.

Admin keys stay unscoped in the dashboard so they can keep managing other API keys. The HTTP API still accepts allowedSessions on any role if you need that from a client.

A session-scoped key still reaches every chat on the sessions it may use. A key can be narrowed further, to chats (a chosen set of groups and individual contacts), with allowedChats on POST /auth/api-keys or PUT /auth/api-keys/{id}. The dashboard does not set or show it yet, and editing a key there leaves its chats unchanged.

• No chats selected — the key can reach every chat on its sessions.
• One or more selected — the key reaches only those chats. Every authenticated REST route not explicitly marked as safe for a chat-scoped key refuses it with 403 (a request naming a session outside allowedSessions still answers 401 first), including routes added in later releases: the refusal is the default. Inside its chats an operator key can do what the marked routes allow, which is more than reading and sending: it can also delete or clear a chat, leave or rename a group, and block the contact. It cannot change who belongs to a group: adding, removing, promoting or demoting participants, answering join requests and reading or resetting the invite link all stay closed.
• The two scopes are independent — a key may be limited to sessions, to chats, to both, or to neither.

This lets you point an AI agent or third-party integration at a shared account without handing it every chat. Give the agent a key scoped to the few groups (or DMs) it is meant to handle: it can send and reply there, but it cannot list your other chats, read any other DM, message a contact outside its set, or reach the queue dashboard. It reads its chats through GET /sessions/{sessionId}/messages/{chatId}/history, which works on whatsapp-web.js only; on Baileys it sees just each chat's last-message preview. It receives no pushed events, so it has to poll. It can still read the session's own status (GET /sessions/{sessionId}) so an integration can tell whether it is connected.

Identity is matched through the lid mapping table: a contact allowlisted by phone number also matches the same person's @lid privacy id once the table maps the two, and an unmapped @lid is refused rather than guessed. A lid's digits are never mistaken for a phone number, so 555000111@lid does not admit [email protected].

The default covers REST routes only. Surfaces that authenticate outside the REST guard do not inherit it, so each one that can return chat data refuses a chat-scoped key with its own check: the /events WebSocket, the MCP mount (per tool call), and the Bull Board queue dashboard. Of the list routes, only GET /sessions/{sessionId}/chats is usable, and it filters to the key's chats before paging.

The API also accepts allowedChats on an admin key, but no admin-only route is open to a chat-scoped key, and the last usable admin key cannot be scoped this way.

None of this changes the ban-risk guidance below. It limits what a key can reach, not what WhatsApp makes of the account.

OpenWA is an unofficial, community-maintained gateway. It connects to WhatsApp through reverse-engineered clients (the whatsapp-web.js project and @whiskeysockets/baileys), not through Meta's official Cloud API. This has real consequences you should understand before you link a phone number.

• There is always a non-zero risk of account restriction or ban. WhatsApp's anti-abuse systems actively look for unofficial automation. No amount of code quality on our side can make that risk zero.

• Pick the right number. Never connect your primary personal or business number to an automated gateway. Use a dedicated number you can afford to lose. If you're running this for paying clients, pass that guidance on to them.

• The two engines trade off differently:

  Engine	Ban-risk profile	Resource cost
  whatsapp-web.js	Lower — drives a real headless Chromium that looks like genuine WhatsApp Web traffic.	High RAM (~300–500 MB / session).
  baileys	Higher — speaks the multi-device WebSocket protocol directly and is easier for WhatsApp to fingerprint.	Low RAM (~30–80 MB / session).

  If account safety is your top priority and you can afford the memory, prefer whatsapp-web.js. If you need density and accept the trade-off, use baileys.

These are practical guardrails, not guarantees — but they materially reduce the chance of WhatsApp flagging the account:

1. Warm up fresh numbers. For the first several days, behave like a normal human user: scan the QR, exchange a handful of messages with saved contacts, join a group or two, set a profile photo. Don't blast on day one.
2. Don't cold-blast strangers. Sending the first-ever message to a large batch of numbers that have never messaged you is the single most reliable way to get restricted — on either engine.
3. Rate-limit yourself. OpenWA ships with a configurable rate limiter (RATE_LIMIT_* env vars). Use it. A few messages per minute per session is sustainable; "thousands in an hour" is not.
4. Use opted-in recipients. The safest workloads are replies and alerts to people who already expect to hear from you (OTP to your own users, order updates, support replies).
5. Keep a fallback. For anything auth-critical or revenue-critical, keep an SMS / email / official-Cloud-API path. Do not bet a login flow solely on an unofficial client.
6. Mind the hosting IP. Cheap datacenter IPs are flagged more aggressively than residential ones. A residential proxy (supported per-session via the proxy settings) can help; it is not a license to spam.

A few things that look like bugs but are actually server-side WhatsApp policy, not OpenWA defects — we track them separately so we can distinguish them from real bugs:

• First message to a brand-new contact sometimes never arrives. The API returns success because the message leaves OpenWA, but WhatsApp's server-side reach-out / trust policy drops it at delivery. This is independent of OpenWA. We track it in #830.
• Accounts that get restricted cannot be "unrestricted" by us. If WhatsApp disables a number, you need to appeal through their channels — OpenWA has no lever to pull.

For any deployment where ethical, legal, or regulatory compliance matters (healthcare, finance, large-scale commercial messaging, anything touching end users in the EU/EEA under DMA/GDPR framings), treat OpenWA as not approved and use Meta's official WhatsApp Cloud API. OpenWA is an excellent fit for personal projects, internal tooling, automation hobbyists, and learning — it is not a drop-in replacement for the official API in regulated environments.

📖 For the deeper, maintainer-side risk analysis (protocol-change exposure, dependency strategy, security posture), see Risk Management (docs/16).

# Clone and start
git clone https://github.com/rmyndharis/OpenWA.git
cd OpenWA
docker compose -f docker-compose.dev.yml up -d

# Access (the dashboard is bundled into the API image and served on the same port)
# Dashboard: http://localhost:2785
# API: http://localhost:2785/api
# Swagger: http://localhost:2785/api/docs

Using Podman instead of Docker? Podman rootless mode requires the socket to be running and DOCKER_HOST to be set:

systemctl --user start podman.socket
systemctl --user enable podman.socket
export DOCKER_HOST=unix:///run/user/$(id -u)/podman/podman.sock

Add the export line to your ~/.bashrc to make it permanent.

# Clone repository
git clone https://github.com/rmyndharis/OpenWA.git
cd OpenWA

# Install the locked dependencies (includes dashboard)
npm ci

# Start API + Dashboard (config is auto-generated on first run)
npm run dev

# Access (in dev the dashboard runs on the Vite server with hot reload)
# Dashboard: http://localhost:2886
# API: http://localhost:2785/api
# Swagger: http://localhost:2785/api/docs

Use npm install instead when intentionally changing dependencies. OpenWA's committed lockfile uses registry artifacts only, so npm 12 works with its secure default that blocks Git dependencies; do not disable that policy globally.

The production stack never exposes /var/run/docker.sock directly to the application container. Instead, a dedicated docker-proxy sidecar (based on tecnativa/docker-socket-proxy) acts as the sole gateway to the Docker daemon:

openwa-api  ──TCP 2375──▶  docker-proxy  ──unix──▶  /var/run/docker.sock

Only the operations needed for container orchestration are enabled (CONTAINERS, IMAGES, VOLUMES, INFO, PING, plus the POST method switch). The application connects via the DOCKER_HOST=tcp://docker-proxy:2375 environment variable, which DockerService detects automatically. Note this is an operational gateway, not a fine-grained privilege boundary: with POST enabled the proxy admits every method to the enabled paths and cannot scope container-create payloads, so a compromised API container would be host-root-equivalent — see SECURITY.md for the full threat model, mitigations, and how to disable the proxy if you don't use the built-in datastore orchestration.

The production image never runs the Node.js process as root. On startup, the container follows this chain:

dumb-init (PID 1)
  └─ docker-entrypoint.sh (root — fixes named-volume ownership via chown)
       └─ gosu openwa node dist/main  (drops to the openwa user)

• dumb-init is PID 1 and forwards signals (SIGTERM, etc.) for graceful shutdown.
• docker-entrypoint.sh runs as root only long enough to chown the named-volume mount points so the openwa user can write to them.
• gosu performs a clean exec-based privilege drop — no su or sudo wrappers, so the node process is the direct child of dumb-init.

Named volumes (e.g. openwa-data) get their ownership corrected automatically on every start, so no manual chown step is needed after volume creation.

For production, use the main docker-compose.yml with optional services:

# Basic production (SQLite, local storage)
docker compose up -d

# With PostgreSQL database
docker compose --profile postgres up -d

# Full stack (PostgreSQL, Redis, MinIO)
docker compose --profile full up -d

The dashboard is bundled into the API image and served by NestJS on the API port, so it needs no profile — it is always available wherever openwa-api runs. For TLS/public exposure, put your own reverse proxy (nginx, Caddy, a cloud load balancer, or a k8s Ingress) in front; see the nginx example in docs/12-troubleshooting-faq.md.

Development vs Production

• Development (docker-compose.dev.yml): SQLite, local storage, API serves the bundled dashboard
• Production (docker-compose.yml): Configurable database, profiles for optional services

Official GHCR images are published as multi-arch manifests for:

• linux/amd64
• linux/arm64

curl -X POST http://localhost:2785/api/sessions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{"name": "my-bot"}'

# Start the session
curl -X POST http://localhost:2785/api/sessions/{sessionId}/start \
  -H "X-API-Key: YOUR_API_KEY"

# Get QR code (scan with WhatsApp)
curl http://localhost:2785/api/sessions/{sessionId}/qr \
  -H "X-API-Key: YOUR_API_KEY"

curl -X POST http://localhost:2785/api/sessions/{sessionId}/messages/send-text \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "chatId": "[email protected]",
    "text": "Hello from OpenWA!"
  }'

curl -X POST http://localhost:2785/api/sessions/{sessionId}/webhooks \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "url": "https://your-server.com/webhook",
    "events": ["message.received", "session.status"],
    "secret": "your-hmac-secret"
  }'

Smart filters (optional): add a filters object to fire the webhook only when conditions match (AND), e.g. { "conditions": [{ "field": "sender", "operator": "is", "value": ["[email protected]"] }] }. Fields: sender / recipient / chatId / body / type / mentions / fromMe / hasMedia / isGroup / kind. A webhook with no filters behaves exactly as before. Use chatId to allowlist specific groups or DMs (e.g. { "field": "chatId", "operator": "is", "value": ["120…@g.us"] }). See the API specification for the full schema.

OpenWA can expose a curated set of tools over the Model Context Protocol so AI agents (Claude, Cursor, …) can drive WhatsApp. It is off by default and additive — every REST route keeps working unchanged.

Set MCP_ENABLED=true to mount a stateless Streamable-HTTP transport at POST /mcp on the existing server (same port, no extra process). It mounts 25 read-only tools by default — session, message, contact, group, webhook, label and automation-rule reads — because the surface is read-only unless you opt out. Add MCP_READONLY=false to mount all 51 tools, adding the write tier (send, reply, group operations). Either way it is a focused surface rather than the full API, so agents aren't overwhelmed.

MCP_ENABLED=true npm run start:prod   # or set MCP_ENABLED in your .env / compose

Point an MCP client at it (e.g. for Claude Code, a .mcp.json at your project root):

{
  "mcpServers": {
    "openwa": {
      "type": "http",
      "url": "http://localhost:2785/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

The key can be passed as Authorization: Bearer … or X-API-Key: …. Every tool call goes through the same API-key auth, role, and per-session scoping as REST.

Security guidance:

• Mint a dedicated, least-privilege key for the agent — a non-admin, session-scoped key (OPERATOR role at most). The plaintext key is shown only once on creation; to rotate, create a new key and delete the old one.
• The key must not carry an IP allow-list (allowedIps) — there is no genuine client IP over MCP, so such a key is rejected.
• Set MCP_READONLY=true to mount only the read tools (no sends/writes).
• Set MCP_RATE_LIMIT_MAX (default 60) to limit tool calls per API key per window.
• Set MCP_RATE_LIMIT_WINDOW_MS (default 60000) to control the sliding window size in milliseconds.
• Do not expose /mcp to the public internet without a fronting auth proxy. For a self-hosted, locally-reached deployment the static API key is appropriate; public exposure should use OAuth 2.1 (not yet built).

openwa/
├── src/
│   ├── main.ts                 # Application entry point
│   ├── app.module.ts           # Root module
│   ├── config/                 # Configuration
│   ├── common/                 # Shared utilities
│   │   ├── cache/              # Redis caching
│   │   └── storage/            # File storage (Local/S3)
│   ├── core/                   # Core systems
│   │   ├── hooks/              # Plugin hooks
│   │   └── plugins/            # Plugin system
│   ├── engine/                 # WhatsApp engine abstraction
│   └── modules/
│       ├── session/            # Session management
│       ├── message/            # Message handling
│       ├── webhook/            # Webhook management
│       ├── group/              # Groups API
│       ├── contact/            # Contacts API
│       ├── auth/               # API key authentication
│       ├── infra/              # Infrastructure management
│       └── health/             # Health checks
├── dashboard/                  # React web dashboard
├── docs/                      # Documentation
├── docker-compose.yml
├── Dockerfile
└── package.json

Comprehensive documentation is available in the docs/ folder:

We welcome contributions! Here's how to get started:

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Please read our Development Guidelines for coding standards and best practices.

This project is licensed under the MIT License – free for personal and commercial use.

See LICENSE for details.