Open SourceSelf-Hostable

OpenClaw was built
for one. ClawScale
is built for everyone.

OpenClaw is one of the most powerful personal AI assistants available. ClawScale takes that same engine and connects it to 12+ chat platforms — adding multi-user management, isolated memory, role-based access, and admin controls. Same power. Built for your whole org, at scale.

ClawScale Architecture
Channels
WhatsApp
Discord
Telegram
Slack
LINE
Teams
Matrix
Signal
WeChat
ClawScale Gateway
AI Backends
OpenClaw #1
OpenClaw #2
GPT-4o
Claude
Custom LLM
Multi-tenant isolation
Many-to-many routing
Per-backend history
The Problem

OpenClaw wasn't built
for a thousand users.

OpenClaw bundles IM gateways and the autonomous agent into one process with shared memory. When hundreds of users interact with the same instance, conversations bleed into each other and memory gets corrupted. You can't scale channels independently, isolate users, or run multiple agent instances.

ClawScale splits these apart — scale gateways and agents independently.


Core Features

User-agent management, not just a gateway.

🔗

12+ messaging channels

WhatsApp, Discord, Telegram, Slack, LINE, Teams, Matrix, Signal, WeChat, and more. Each channel has a dedicated adapter with platform-native features.

🤖

Pluggable AI backends

Route messages to OpenAI, Anthropic, OpenRouter, OpenClaw instances, or any OpenAI-compatible endpoint. Run multiple backends per tenant simultaneously.

🔄

Many-to-many routing

End-users can talk to multiple agents at once. Messages route to all active backends and replies are labeled by source. Users manage their agent team via slash commands.

🏢

Multi-tenant isolation

Every user's conversations, memory, and state are fully isolated. Hundreds or thousands of users interact with agents without corrupting each other's context. Data never crosses boundaries.

🛡️

Admin dashboard

Web UI for managing channels, AI backends, conversations, team members, workspace settings, and audit logs. Role-based access: Admin, Member, Viewer.

⚖️

Horizontal scaling

Scale OpenClaw instances (or other agents) independently behind one gateway. Add channels without touching agent config. Add agents without touching channel config.

How It Works

Messages in, replies out.

01

End-user sends a message

A user messages your bot on WhatsApp, Discord, Telegram, or any connected channel. The channel adapter normalizes the message and posts it to the ClawScale gateway.

02

Gateway routes to backends

ClawScale identifies the user, loads conversation history, parses any commands, and routes the message to the user's active AI backend(s). Multiple backends can respond in parallel.

03

AI responds via the channel

Each backend generates a reply with its own conversation context. Replies are saved, labeled by backend, and delivered back through the original channel. The user sees a natural conversation.

Channels

Deploy across 12+ platforms

Each channel has a dedicated adapter. Connect from the dashboard, provide credentials, and go live.

WhatsApp (Personal)

QR code scan via Baileys

WhatsApp Business

Meta Cloud API webhook

Discord

Bot token (discord.js)

Telegram

Bot token (grammy long-polling)

Slack

Bot + App-level token (Socket Mode)

LINE

Channel access token + secret (webhook)

Signal

signal-cli REST API

Microsoft Teams

Azure Bot Service (webhook)

Matrix

Homeserver URL + access token

WeChat Work (WeCom)

Bot token (WebSocket)

WeChat Personal

QR code scan (iLink Bot API)

Web Chat Widget

Webhook integration

AI Backends

Plug in any agent.

ClawScale never injects hidden prompts. Each backend manages its own system prompt, model, and tools. Scale from one OpenClaw instance to many, or mix in other providers.

OpenClaw

Route to one or more OpenClaw instances. OpenClaw manages its own tools, memory, and prompts.

OpenAI

GPT models via OpenAI API. Configure model, system prompt, and API key per backend.

Anthropic

Claude models via Anthropic API. Full system prompt and model control.

OpenRouter

Multi-model aggregator. Access hundreds of models through one API key.

Pulse

Pulse Editor AI agent. Connect to your Pulse instance for coding-aware responses.

Custom

Any OpenAI-compatible endpoint. Self-hosted models, vLLM, Ollama, or any proxy.

Comparison

OpenClaw vs ClawScale

OpenClaw bundles gateways and agent into one process. ClawScale decouples them so each can scale independently.

FeatureOpenClawClawScale
ArchitectureMonolithicDecoupled gateway + agents
IM channels
Autonomous agentVia OpenClaw backend
User isolationShared memoryPer-user/per-tenant
Scale to 1000s of usersMemory corruption risk
Multiple agent instances
Pluggable AI backends
Many-to-many user/agent
Admin dashboard
Role-based access
Audit logs
LicenseOpen sourceOpen source

Both projects are open source. ClawScale uses OpenClaw as one of its pluggable AI backends.

From Users

Teams running ClawScale

We were running three separate OpenClaw instances for different teams. ClawScale unified everything behind one gateway with proper tenant isolation.

Wei Ling T.
Operations Lead, 30-person trading firm

Being able to plug in both GPT-4o and Claude as backends, then let users pick which agent to talk to, changed how our support team works.

Reza A.
Founder, 15-person logistics startup

We connected WhatsApp, Discord, and Telegram to a single ClawScale instance. One dashboard to manage all channels and see every conversation.

Joanne C.
Head of Customer Success, SaaS company
Pricing

Start free. Scale when you're ready.

Self-host for free, or use managed ClawScale for your team.

Starter
Free
  • Up to 5 users
  • 1 persona
  • 1GB knowledge base
Get started free
Most Popular
Business
$49
/month
  • Up to 50 users
  • 5 personas
  • 10GB knowledge base
Start Business trial
Enterprise
Contact us
  • Unlimited users
  • Custom personas
  • Unlimited knowledge base
Talk to sales
Get Started

Deploy once. Connect every channel and agent.

Open source, self-hostable, and ready to run. Clone the repo and have your gateway live in minutes.

MIT licensed Self-hostable 12+ channels 6 AI backend types