OpenClaw, but in containers: Meet NanoClaw

"NanoClaw looks really interesting in that the core engine is ~4000 lines of code (fits into both my head and that of AI agents, so it feels manageable, auditable, flexible, etc.) and runs everything in containers by default. I also love their approach to configurability - it's not done via config files, it's done via skills!"

"They're running bare metal with some application level checks to try to prevent it from accessing things it shouldn't access. NanoClaw, each agent runs in its own container. And that's really important. If you take the whole instance of OpenClaw and put it in a container, that doesn't really help you because you've connected it to so many different things that are all in that container with the agent."

OpenClaw, a popular open-source agent platform, has experienced security incidents as its usage increased, including incidents where agents caused unintended damage like deleting user inboxes. This vulnerability demonstrates the risks of unrestrained AI agents. Gavriel Cohen developed NanoClaw, a more secure alternative that constrains agent behavior through containerization and a compact codebase of approximately 4,000 lines. Unlike OpenClaw's bare-metal approach with application-level checks, NanoClaw isolates each agent in its own container, preventing widespread system access. The platform's small, auditable codebase and skill-based configurability have attracted attention from prominent AI researchers, including Andrej Karpathy, who praised its manageability and flexibility.

Read at Theregister