You've seen OpenClaw — the open-source AI assistant with 340K+ GitHub stars that connects to 20+ messaging platforms. You want to run it yourself. Your data stays on your machine, you pick your own LLM, no monthly fees.
Then you hit the wall.
Docker setup, port mapping, environment variables, supervisord configs, getting the gateway process to actually start — and if something breaks, you're debugging inside a container with no GUI. Roughly 1 in 5 people who attempt self-hosting never get a working instance. The deployment friction kills the experience before it begins.
ClawFleet eliminates that entire wall. One command installs everything. A browser dashboard replaces the CLI. Ten minutes from now, your first AI agent can be live.
Install
curl -fsSL https://clawfleet.io/install.sh | sh
This single command handles everything:
- Installs Docker if you don't have it (Colima on macOS, Docker Engine on Linux)
- Downloads the
clawfleetCLI - Pulls the pre-built sandbox image (~1.4 GB)
- Starts a web dashboard at http://localhost:8080
No YAML files to write. No docker-compose to debug. No manual config.
Set Up Your First Agent (5 minutes)
The installer opens a dashboard in your browser. From there, three steps to a running agent:
1. Register a model — Assets → Models. Paste an API key from OpenAI, Anthropic, Google, or DeepSeek. ClawFleet validates the key before saving — you'll know immediately if something's wrong.
2. Create an instance — Fleet → Create. Each instance is a fully isolated Docker container with its own XFCE desktop, accessible through your browser via noVNC. Your host system is completely untouched.
3. Assign and launch — Pick the model you just registered, assign it to the instance, click "Restart Bot". That's the entire configuration.
Your agent is now live. It's running on your hardware, your data stays local, and there's no subscription. If you close the laptop and reopen it, the agent auto-recovers — no manual restart needed.
Beyond the First Bot
Once your first agent is running, the dashboard unlocks everything else without touching the terminal:
Character system — Define a persona with a bio, backstory, and communication style. Assign it to an instance and the agent takes on that personality across all conversations. Think of it as a job description for an AI employee.
Messaging channels — Connect Telegram, Discord, Slack, or Lark. Your agent responds to real users on real platforms. Each channel is validated before saving so you don't deploy a broken bot.
Skills marketplace — 52 built-in skills (weather, GitHub, coding, web search) plus 13,000+ community skills on ClawHub. Browse and install from the dashboard — one click, no CLI.
Fleet scaling — Need multiple agents? Create more instances, each with a different model, personality, and channel. On a 16GB Mac you can comfortably run 3 bots simultaneously. They're aware of each other and collaborate naturally — an engineer bot will @-mention your marketing bot when a question crosses domains.
Soul Archive — When a bot is performing well, snapshot its entire state (personality, memory, config). Clone it to new instances instantly. No retraining.
Resource Usage
Tested on M4 MacBook Air (16 GB RAM):
| Instances | RAM (idle) |
|---|---|
| 1 | ~1.5 GB |
| 3 | ~4.5 GB |
Lightweight enough to run alongside your daily work.
The Cost
| Managed hosting (2 bots) | ClawFleet (3 bots) | |
|---|---|---|
| Subscription | ~$20/month | $0 |
| LLM API tokens | included | ~$25/month |
| Extra hardware | N/A | $0 (your existing Mac) |
You trade a subscription for direct API access — and gain a third bot, full data ownership, and zero vendor lock-in.
Get Started
curl -fsSL https://clawfleet.io/install.sh | sh
Ten minutes from now you'll have a self-hosted AI agent running on your own machine. The wiki has step-by-step guides for every LLM provider and messaging platform if you need them.
Star ClawFleet on GitHub — it helps others find the project.
Join the Discord — see what others are building with their fleets and get help if you're stuck.
Open source. MIT licensed. Free forever.
Top comments (0)