nanoclaw

Why we included this project

NanoClaw is for people who want an AI assistant that can actually act in their chat apps, replying from WhatsApp, Telegram, Slack, or Discord, keeping memory across conversations, and running scheduled jobs, but who are not comfortable handing a black box the keys to their machine. The codebase is deliberately small, one process and a handful of files, so a developer can read it and modify it rather than trust it blindly. The security model is the interesting part: each agent runs inside its own Linux container with filesystem isolation, so the OS bounds what the assistant can reach instead of relying on permission checks inside the app. It builds on Anthropic's Agents SDK, and the setup script walks you from a fresh machine to a named agent you can message, even calling in Claude Code to diagnose and resume if a step fails. That combination of a small, auditable codebase and real OS-level isolation is what makes it worth a look for individuals and small teams who want to self-host their assistant.