🌐 Website: clawfleet.io · 💬 Community: Discord · 📝 Blog: Dev.to
Deploy and manage a fleet of isolated OpenClaw instances on a single machine — each sandboxed in Docker, managed from a browser dashboard.
curl -fsSL https://clawfleet.io/install.sh | sh10 minutes: Docker installed, image pulled, dashboard running at http://localhost:8080. Log in with your ChatGPT account — your existing Plus subscription covers inference, no API keys needed.
Imagine buying N dedicated Mac Minis, each running its own OpenClaw instance, fully isolated, collaborating in Discord. Your own AI company — data stays on your hardware, no SaaS subscription.
ClawFleet makes that free. Each instance runs in its own Docker container with isolated filesystem and networking. On your existing Mac or Linux box. ~1.5 GB RAM per instance.
- Sandboxed instances — each OpenClaw runs in its own Docker container, isolated from your host and from each other. No rogue skill can read your files
- Browser dashboard — create, configure, monitor, and destroy instances without touching a terminal
- ChatGPT login — authenticate with your existing ChatGPT account, or use API keys from OpenAI, Anthropic, Google, DeepSeek
- Version pinning — lock a tested OpenClaw version so upstream breaking changes don't touch you
- Fleet management — spin up as many instances as your RAM allows, each with different models, personas, and channels
- Character system — define reusable personas (bio, backstory, style, traits) and assign them to instances
- Skill management — browse 52 built-in skills, search and install from 13,000+ community skills on ClawHub
- Full desktop per instance — each claw has an XFCE desktop accessible via noVNC in your browser
- Soul Archive — save a configured instance's soul and clone it instantly
- Auto-recovery — instances automatically restart their gateway after container restarts
- macOS or Linux
The install command above will:
- Install Docker if needed (Colima on macOS, Docker Engine on Linux)
- Download and install the
clawfleetCLI - Pull the pre-built sandbox image (~1.4 GB)
- Start the Dashboard as a background daemon
- Open http://localhost:8080 in your browser
Linux server deployment notes
The Dashboard listens on all interfaces (0.0.0.0:8080) by default on Linux, so you can access it remotely at http://<server-ip>:8080. To restrict to localhost only:
clawfleet dashboard stop
clawfleet dashboard start --host 127.0.0.1To access the Dashboard from your local machine via SSH tunnel:
ssh -fNL 8081:127.0.0.1:8080 user@your-server
# Then open http://localhost:8081 in your browser
# To stop the tunnel later: kill $(lsof -ti:8081)The -fN flags run the tunnel in the background so you can close your terminal without breaking the connection. Port 8081 is used here because 8080 is often occupied by a local ClawFleet instance.
The Control Panel (OpenClaw's built-in web UI) requires a secure context for WebSocket device identity — the SSH tunnel provides this. All other Dashboard features (fleet management, configuration, Restart Bot, etc.) work without a tunnel via direct HTTP.
Manual install? See the Getting Started wiki page.
Think of ClawFleet as your AI company. Assets are the tools and resources your company owns; Fleet is your team of AI employees. You assign different tools to different employees, and put your AI workforce into production.
Assets → Models — register LLM API keys. These are the "brains" your employees think with. Each model is validated before saving.
Assets → Characters — define reusable personas. Think of them as "job descriptions" — Tony Stark the CTO, Steve Jobs the CPO, Ray Kroc the CMO. Give each character a bio, backstory, communication style, and personality traits.
Assets → Channels — connect messaging platforms (Telegram, Discord, Slack, etc.). These are the "workstations" where your employees serve customers. Optional; validated before saving.
Fleet → Create — spin up OpenClaw instances. Each one is a new employee joining your company.
Fleet → Configure — assign a model, character, and channel to each instance. Give your CTO a Claude brain and a Discord workstation. Give your CMO a GPT brain and a Slack feed. Different employees, different tools, different personalities.
Fleet → Skills — each instance has access to 52 built-in skills (weather, GitHub, coding, and more). Want more? Search 13,000+ community skills on ClawHub and install them with one click. Different employees can learn different skills.
Once an employee is trained and performing well, save their soul — personality, memory, model config, and conversation history — so you can clone them instantly.
Fleet → Save Soul — click on any configured instance to save its soul to the archive.
Fleet → Soul Archive — browse all saved souls, ready to be loaded into new hires.
Fleet → Create → Load Soul — when creating new instances, pick a soul from the archive. The new employee starts with all the knowledge and personality of the original — no retraining needed.
Click "Desktop" on any running instance to open its detail page — embedded noVNC desktop, live logs, and real-time resource charts.
Connect your fleet to messaging platforms and watch your AI employees work together. Here, an engineer, product manager, and marketer welcome a new teammate — all running autonomously in a Discord group chat.
See the Wiki for full documentation, including:
- Getting Started — prerequisites, install, first instance
- Dashboard Guide — sidebar navigation, asset management, fleet management
- LLM Provider guides — Anthropic | OpenAI | Google | DeepSeek
- Channel guides — Telegram | Discord | Slack | Lark
- CLI Reference | FAQ
Every command supports --help for detailed usage and examples:
clawfleet --help # List all available commands
clawfleet dashboard --help # Show dashboard subcommandsQuick reference:
clawfleet create <N> # Create N claw instances (image must be pre-built)
clawfleet create <N> --pull # Create N instances, pull image from registry if missing
clawfleet configure <name> # Configure an instance with a model and optional channel credentials
clawfleet list # List all instances and their status
clawfleet desktop <name> # Open an instance's desktop in the browser
clawfleet start <name|all> # Start a stopped instance
clawfleet stop <name|all> # Stop a running instance
clawfleet restart <name|all> # Restart an instance (stop + start)
clawfleet logs <name> [-f] # View instance logs
clawfleet destroy <name|all> # Destroy instance (data kept by default)
clawfleet destroy --purge <name|all> # Destroy instance and delete its data
clawfleet snapshot save <name> # Save an instance's soul to the archive
clawfleet snapshot list # List all saved souls
clawfleet snapshot delete <name> # Delete a saved soul
clawfleet create 1 --from-snapshot <soul> # Create instance from a saved soul
clawfleet dashboard serve # Start the Web Dashboard
clawfleet dashboard stop # Stop the Web Dashboard
clawfleet dashboard restart # Restart the Web Dashboard
clawfleet dashboard open # Open the Dashboard in your browser
clawfleet build # Build image locally (offline/custom use)
clawfleet config # Show current configuration
clawfleet version # Print version infoTo destroy all instances (including data), stop the Dashboard, and remove all build artifacts — effectively returning to a clean slate:
make resetAfter resetting, start over from Quick Start step 1.
Tested on M4 MacBook Air (16 GB RAM):
| Instances | RAM (idle) | RAM (Chromium active) |
|---|---|---|
| 1 | ~1.5 GB | ~3 GB |
| 3 | ~4.5 GB | ~9 GB |
| 5 | ~7.5 GB | not recommended |
Actively developed. Both CLI and Web Dashboard are functional.
Contributions and feedback welcome — please open an issue or PR.
If you run into any problems, feel free to reach out: weiyong1024@gmail.com
MIT