Architecture

What Is Docker Sandbox?

A Docker container configured with restricted permissions that isolates an AI agent from the host system and other containers.

In Plain English

A Docker sandbox wraps your AI agent in a container with strict constraints: limited file system access, controlled network, no host system access, and restricted process capabilities. The agent can do its job but cannot escape its container.

OpenClaw uses Docker-in-Docker (DinD) for its sandbox architecture. Each tenant gets an isolated Docker environment where their agent runs. A Docker socket proxy adds another layer — the agent can manage its own containers but cannot access other tenants' resources.

This architecture means even if an agent achieves code execution through an MCP tool, the damage is contained to its own sandbox.

Why It Matters for OpenClaw

Docker sandboxing is the industry standard for containing untrusted workloads. It provides strong isolation without the overhead of full virtual machines, making it practical for per-agent deployment.

How Clawctl Helps

Clawctl deploys each agent in a Docker sandbox with a socket proxy sidecar, health checks, and auto-recovery. The sandbox is configured automatically — no Docker expertise required.

Try Clawctl — 60 Second Deploy

Common Questions

Is Docker sandboxing secure enough?

For most workloads, yes. Clawctl adds socket proxy filtering, network policies, and egress controls for defense-in-depth.

Can I customize the sandbox?

Enterprise plans support custom Docker configurations. Standard plans use Clawctl's hardened default configuration.

What about Kubernetes?

Clawctl supports both Docker and Kubernetes deployments. K8s adds pod-level isolation and network policies.