Deploy and manage a fleet of isolated OpenClaw instances on a single machine — from a browser dashboard, no CLI needed.
You don't need a dedicated server. If you have a Mac with Apple Silicon, ClawSandbox lets you:
- Deploy OpenClaw in minutes — fully sandboxed in Docker, completely isolated from everything else on your machine
- Run as many as you want — spin up an entire fleet of OpenClaw instances and experience a one-person company powered by AI
No cloud bills. No new hardware. Everything runs on the machine you already have.
LLM AI applications are evolving through three stages:
- ChatBot — helps everyone access knowledge
- Agent — makes everyone a professional
- OpenClaw — makes everyone a manager
OpenClaw is a self-hosted personal AI assistant that connects to 20+ messaging platforms including WhatsApp, Telegram, and Slack. ClawSandbox removes the deployment bottleneck — instead of struggling to run a single instance, you can spin up an entire fleet with one command.
- One-command fleet deployment — give it a number, get that many isolated OpenClaw instances
- Web Dashboard — manage your entire fleet from a browser with real-time stats, one-click actions, and embedded noVNC desktops
- Full desktop per instance — each claw runs in its own Docker container with an XFCE desktop, accessible via noVNC
- Lifecycle management — create, start, stop, restart, and destroy instances via CLI or Dashboard
- Soul Archive — save a configured instance's soul (personality, memory, model config) and clone it to new instances instantly
- Auto-recovery — configured instances automatically restart their gateway after container restarts
- Data persistence — each instance's data survives container restarts
- Resource isolation — instances are isolated from your host system and from each other
- macOS or Linux
- A Docker environment (e.g. Docker Desktop)
git clone https://github.com/weiyong1024/ClawSandbox.git
cd ClawSandbox
make build
# If Go is missing, ClawSandbox bootstraps the Go version from go.mod
# into a user-local toolchain directory automatically.
# Optionally install to PATH (otherwise use ./bin/clawsandbox in place of clawsandbox below):
sudo make installclawsandbox dashboard serve
# Open http://localhost:8080Click "System → Image" in the Dashboard and build the sandbox image (~1.4 GB, first build takes several minutes).
Think of ClawSandbox 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 → 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 and channel from your asset pool to each instance. Different employees can use different tools for different jobs.
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:
clawsandbox --help # List all available commands
clawsandbox dashboard --help # Show dashboard subcommandsQuick reference:
clawsandbox create <N> # Create N claw instances (image must be pre-built)
clawsandbox create <N> --pull # Create N instances, pull image from registry if missing
clawsandbox list # List all instances and their status
clawsandbox desktop <name> # Open an instance's desktop in the browser
clawsandbox start <name|all> # Start a stopped instance
clawsandbox stop <name|all> # Stop a running instance
clawsandbox restart <name|all> # Restart an instance (stop + start)
clawsandbox logs <name> [-f] # View instance logs
clawsandbox destroy <name|all> # Destroy instance (data kept by default)
clawsandbox destroy --purge <name|all> # Destroy instance and delete its data
clawsandbox snapshot save <name> # Save an instance's soul to the archive
clawsandbox snapshot list # List all saved souls
clawsandbox snapshot delete <name> # Delete a saved soul
clawsandbox create 1 --from-snapshot <soul> # Create instance from a saved soul
clawsandbox dashboard serve # Start the Web Dashboard
clawsandbox dashboard stop # Stop the Web Dashboard
clawsandbox dashboard restart # Restart the Web Dashboard
clawsandbox dashboard open # Open the Dashboard in your browser
clawsandbox build # Build image locally (offline/custom use)
clawsandbox config # Show current configuration
clawsandbox 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