ZeroClaw is open-source infrastructure for autonomous personal AI assistants, with 33,000+ stars on GitHub — fast, small and security-first, written in Rust.
Personal agents that can act on your behalf are only trustworthy if their autonomy is bounded. ZeroClaw builds that boundary into the infrastructure: by default an agent runs “supervised” — medium-risk operations need your approval and high-risk ones are blocked — with workspace boundaries, command policies and OS-level sandboxes containing what it can touch. On top of that safety core, ZeroClaw provides the runtime for a fully autonomous personal assistant that can connect to your channels and tools, deploy on any operating system or platform, and swap models and components freely. Being written in Rust, the whole thing is a small, fast binary rather than a heavyweight stack. The project is Apache-2.0 licensed and under very active development, updated in October 2026.
ZeroClaw belongs to the fast-growing ‘claw’ family of personal-agent systems — but where many rivals optimize for capability first, it argues the winning design is the one a cautious user can actually trust with real accounts and files. Escape hatches matter as much as the defaults: experts can loosen supervision deliberately, per action class, instead of choosing between a toy and a liability. Its small footprint also makes it practical on modest hardware — a spare mini-PC or a cheap VPS can host an always-on assistant without a GPU or a cloud bill beyond model tokens.
ZeroClaw is free and open source under the Apache License 2.0. Your only costs are hosting and the model APIs you connect.
Developers and power users who want a personal AI assistant with real capabilities and are not willing to give an agent unbounded access to their machine.