NanoClaw (qwibitai/nanoclaw) is a security-focused lightweight alternative to OpenClaw with approximately 21,500 GitHub stars. Written in TypeScript, it consists of only ~700 lines of code.
All agents run in isolated Linux containers — Apple Container on macOS or Docker on other platforms — ensuring that bash access runs inside the container and not on the host system. The binary footprint is approximately 800 KB with a minimum of 64 MB RAM for container isolation.
NanoClaw connects to WhatsApp, Telegram, Slack, Discord, and Gmail. It is built on the Anthropic Agents SDK. Supports Linux, macOS (Apple Container), and Windows (Docker).
Frequently Asked Questions
What makes NanoClaw different from OpenClaw?
NanoClaw takes a security-first approach by running every agent inside isolated Linux containers (Apple Container on macOS, Docker elsewhere). This means bash and file access happen inside the container, not on your host system. OpenClaw runs directly on the host, which is more convenient but carries more risk if a skill misbehaves. NanoClaw is also significantly smaller at ~800 KB vs OpenClaw’s npm installation.
Is NanoClaw related to HKUDS nanobot?
These are two different projects with similar names. NanoClaw (qwibitai/nanoclaw) is the security-focused containerized implementation. HKUDS nanobot (from the Hong Kong University Data Intelligence Lab) is a research-oriented lightweight implementation. ClawTrackr tracks both separately — check the HKUDS nanobot page for the HKU project.
How much RAM does NanoClaw use?
NanoClaw itself is lightweight, but the container isolation requires a minimum of 64 MB RAM for the container runtime (Apple Container or Docker). This is still dramatically less than OpenClaw’s ~390 MB footprint, and the security tradeoff is worth it for many deployments.
Does NanoClaw support the same messaging platforms as OpenClaw?
NanoClaw currently supports WhatsApp, Telegram, Slack, Discord, and Gmail. It doesn’t yet have the 50+ platform coverage of OpenClaw, but covers the most common platforms. It’s built on the Anthropic Agents SDK and targets Claude-first workflows.