Short answer: Yes. Long answer: You're in control of your data, and here's exactly how.
100% of the code is public and auditable
Runs on your machine, not our servers
We don't collect or track anything
You use your own API credentials
When self-hosted, OpenClaw runs entirely on your machine. Your data never passes through third-party servers (except your chosen AI provider).
Every line of code is public on GitHub. Security researchers, developers, and the community continuously audit the codebase.
You use your own API keys directly with AI providers. We never see or store your API credentials.
OpenClaw doesn't phone home. No analytics, no tracking, no hidden data collection—unless you explicitly enable it.
Your conversations, preferences, and memory files stay on your machine. Back them up, encrypt them, or delete them anytime.
Fine-grained control over what OpenClaw can access—files, network, tools—all configurable in your config file.
Here's exactly what happens when you send a message:
The only external service that sees your messages is your AI provider (OpenAI, Anthropic, etc.). OpenClaw has no servers that process your data. When you self-host, we literally cannot see what you're doing.
For maximum privacy, run OpenClaw with local AI models. Your data never leaves your machine—not even to an AI provider.
# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh
# Pull a model
ollama pull llama3
# Configure OpenClaw to use it
openclaw config set model.provider ollama
openclaw config set model.name llama3
# Now everything runs locally! 🎉Never put API keys directly in config files that might be shared or committed to git.
export OPENAI_API_KEY="sk-..."Only grant access to directories OpenClaw actually needs.
security:
file_access:
- ~/.openclaw/workspace
- ~/Documents/notesIf exposing the gateway to the internet, always enable authentication.
gateway:
auth:
enabled: true
token: "your-secure-token"Running in a container adds an extra security layer.
docker run -d --name openclaw \
-v openclaw-config:/home/openclaw/.openclaw \
openclaw/openclaw:latest| Aspect | OpenClaw | ChatGPT | Google Gemini |
|---|---|---|---|
| Open source | ✅ Yes | ❌ No | ❌ No |
| Self-host option | ✅ Yes | ❌ No | ❌ No |
| Local AI models | ✅ Yes | ❌ No | ❌ No |
| Your data on your servers | ✅ Yes | ❌ No | ❌ No |
| No telemetry | ✅ Yes | ⚠️ Tracked | ⚠️ Tracked |
| Data deletion control | ✅ Full | ⚠️ Request | ⚠️ Request |
| Code auditable | ✅ Yes | ❌ No | ❌ No |
Only if you explicitly grant file access in your configuration. By default, OpenClaw can only access its own workspace directory. You control exactly which folders and files it can see.
Your messages are sent only to your configured AI provider (like OpenAI or Anthropic) to generate responses. OpenClaw itself has no servers that collect your data. We don't see your conversations.
That depends on your AI provider's policies. OpenAI and Anthropic both offer API usage that isn't used for training. Check your provider's data policy. OpenClaw gives you the choice—you can even run fully local models with Ollama.
Since OpenClaw runs on your machine, you're in control of security. We recommend: keeping software updated, using strong API keys, limiting file permissions, and running in a container if you want extra isolation.
Yes! Use local AI models via Ollama (like Llama, Mistral, or Phi). Your data never leaves your machine. Performance depends on your hardware.
Our managed service uses industry-standard security: encrypted connections, isolated containers, EU/US data centers with compliance certifications. We still never train on your data, and you can delete everything anytime.
The open-source codebase is continuously reviewed by the community. We welcome security researchers—see our SECURITY.md for responsible disclosure guidelines.
Start using an AI assistant that respects your privacy.