Connect your AI agents to any messaging platform

ClawScale is an open-source gateway that connects AI agents — OpenClaw, Claude Code, LLMs, or any custom agent — to WhatsApp, Discord, Slack, Telegram, and 10+ other messaging platforms. It handles multi-tenant isolation, conversation routing, and backend orchestration so hundreds of users can talk to your agents without interfering with each other.

• Deploy AI agents to messaging platforms — connect any LLM or AI agent to WhatsApp, Discord, Slack, Telegram, Teams, and more from a single dashboard
• Support many users at once — each user gets isolated conversations, memory, and state. No cross-contamination between users
• Mix and match AI backends — run OpenClaw, GPT, Claude, self-hosted models, or all of them at once. Users can talk to multiple agents in the same chat
• Manage everything from a dashboard — channels, backends, users, roles, and audit logs in one place

Add channels from the dashboard — provide credentials and hit Connect. WhatsApp and WeChat Personal show a QR code for pairing.

ClawScale doesn't lock you into one AI provider. Connect any combination of these:

Users can have multiple backends active at once. Replies are labeled by source (e.g. [GPT-4o], [Claude]) so users know which agent is responding.

• Node.js 20+
• pnpm
• Docker (for PostgreSQL)

# 1. Start Postgres
docker compose up postgres -d

# 2. Configure environment
cp .env.example .env
# Edit .env — defaults work for local dev

# 3. Install dependencies
pnpm install

# 4. Push database schema
cd packages/api && pnpm db:push && cd ../..

# 5. Start API and web (separate terminals)
cd packages/api && pnpm dev
cd packages/web && pnpm dev

• Dashboard: http://localhost:4040
• API: http://localhost:4041

Or run everything with Docker:

cp .env.example .env
docker compose up --build

Open the dashboard and Register to create your workspace. You're the admin.

End-user (WhatsApp, Discord, Slack, etc.)
    |
    v
Channel Adapter ──> POST /gateway/:channelId
    |
    v
Message Router
    ├── Parse commands (/team, /backends, etc.)
    ├── Resolve target backend(s)
    ├── Load conversation history
    ├── Call AI backend(s)
    └── Save messages + return reply
    |
    v
Channel Adapter ──> Reply to end-user

1. A user sends a message on any connected platform
2. The channel adapter normalizes the message and forwards it to ClawScale
3. ClawScale routes the message to the right AI backend(s), keeping each user's conversation history isolated
4. The AI response is sent back through the same channel

End-users can run commands directly in chat to manage their experience:

To message a specific backend: gpt> explain quantum computing

ClawScale is designed for multi-user deployments. Every user's conversations, memory, and state are fully isolated — data never crosses boundaries.

Access control — workspace admins decide who can interact with the bot:

• Anonymous — anyone can chat (default)
• Whitelist — only approved users
• Blacklist — block specific users

Roles — each workspace has three roles:

Plans: Starter (5 members, 3 channels), Business (50 members, 20 channels), Enterprise (unlimited).

ClawScale originated from OpenClaw. OpenClaw bundles messaging gateways and an AI agent into one process — great for personal use, but conversations bleed into each other when multiple users share the same instance.

ClawScale separates the gateway layer from the agent layer, so each can scale independently:

• API — Hono + Prisma + PostgreSQL
• Web — Next.js 16 + React 19 + Tailwind CSS
• AI — OpenAI SDK, Anthropic SDK, LangChain
• Monorepo — pnpm workspaces

packages/
├── api/       # Backend, channel adapters, AI routing, Prisma schema
├── web/       # Next.js dashboard
└── shared/    # Shared TypeScript types

MIT