Security Architecture
OpenClaw is built on a "Trust No One" architecture. From sandboxed execution to local-first memory, every layer is designed to keep your business safe.
The Sandbox:
A Safe Playground for AI
We don't let AI run wild on your computer. OpenClaw uses Docker containers to create isolated "sandboxes" for executing tasks.
Isolated Execution
When the AI writes code or runs a command, it happens inside a disposable container. If something goes wrong, it only affects the sandbox, not your real files.
Simulation Mode
The AI can "simulate" an action first. You can see exactly what it plans to do (e.g., "Delete these 5 files") and approve it before it actually happens.
> Mounting temp volume...
> Running cleanup script...
Action blocked by sandbox policy.
Where Does My Data Go?
Unlike cloud-only chatbots, OpenClaw treats your hard drive as the source of truth.
Local Memory Files
The AI's "brain" is just a set of Markdown files on your computer. You can open them, read them, and edit them. It's not a black box database.
The Gateway
All messages pass through a central Gateway. This acts as a firewall, filtering out unauthorized users and ensuring only approved commands reach the AI.
API Privacy Control
When we do use cloud AI (like Claude), we can sanitize data first. You choose exactly what information is allowed to leave your network.