도구 소개
Introduction
What is NanoClaw?
NanoClaw is a radically simple and secure AI assistant built around Anthropic's Claude. It is designed to be a lightweight, understandable, and fully customizable alternative to complex AI agent frameworks.
Core Philosophy
NanoClaw champions security through simplicity and isolation. Instead of relying on intricate application-level permission systems, it leverages real OS-level container isolation (via Apple Container on macOS or Docker on Linux) to sandbox each chat session, ensuring commands run safely inside a container, not on your host system.
Designed for Customization
This is not a bloated framework but working software intended to be forked. With a core codebase of only ~500 lines of TypeScript, it is small enough to be read and understood in under 10 minutes, empowering users to directly modify the code to match their exact requirements without configuration sprawl.
Features
Essential Built-in Capabilities
NanoClaw comes with a robust set of core features to get you started immediately, complemented by a modular skills system for extended functionality.
WhatsApp Integration
Message your Claude assistant directly from your phone using the integrated WhatsApp interface, providing a convenient and accessible chat front-end.
Container Isolation & Isolated Groups
Each chat session runs in its own sandboxed Linux container, providing strong security boundaries. Furthermore, each chat group maintains its own isolated memory and filesystem context.
Scheduled Tasks
Set up automated, recurring jobs that run Claude on a schedule. Perfect for daily briefings, weekly reports, or any routine AI-powered task.
Web Access & Skills System
The assistant includes built-in web search and content fetching capabilities. Optional integrations (like Gmail) can be added via the pluggable skills system, allowing you to extend functionality as needed.
Frequently Asked Questions
Why use WhatsApp and not another messaging app?
The author built it for personal use with WhatsApp. The entire point of NanoClaw is customization—if you prefer Telegram or Signal, you can fork the repository and modify the code to change it. The codebase is intentionally small to make such changes safe and straightforward.
Is NanoClaw secure?
Yes. Agents run in genuine Linux containers with only explicitly mounted directories accessible. This provides a stronger security model than typical application-level permission checks. However, users should still exercise caution and review any code they run, which is feasible given the project's small size.
Can I run NanoClaw on Linux?
Absolutely. During the /setup process, NanoClaw will automatically detect the Linux environment and configure Docker as the container runtime.
Why are there no configuration files?
NanoClaw avoids configuration sprawl by encouraging direct code modification. The philosophy is that every user should tailor the software to their precise needs by changing the source code, resulting in a system that does exactly what you want, not what a generic configuration allows.
How do I debug issues?
Use an AI-native approach: ask Claude Code. You can prompt it with questions like "Why isn't the scheduler running?", "What's in the recent logs?", or "Why did this message not get a response?" to diagnose problems.
What are the requirements?
You need an operating system (macOS with Apple Silicon recommended, or Linux x86_64/ARM64), Node.js 20+, Claude Code, and either Apple Container (on macOS) or Docker (on Linux or macOS).