NanoClaw is a lightweight, open-source personal AI assistant built on the Anthropic Claude Agent SDK. Agents run in isolated Linux containers (Apple Container on macOS, Docker on macOS/Linux). WhatsApp built-in, other channels via skills. Features persistent SQLite memory, scheduled tasks, web access, browser automation, and Agent Swarms.

Run 'git clone https://github.com/qwibitai/NanoClaw.git && cd NanoClaw && claude' and use /setup in the Claude session. Claude Code handles dependencies, authentication, container setup, and service configuration. Requires Node.js 20+ and Claude Code.

NanoClaw is small enough to understand — one process, a few source files, no microservices. Unlike bloated frameworks, you can read the entire codebase. Agents run in real Linux containers with filesystem isolation, not behind application-level permission checks.

NanoClaw runs exclusively on the Anthropic Claude Agent SDK. Claude Code is installed globally inside each container and handles all AI interactions. The SDK manages model selection internally — NanoClaw is single-provider by design.

WhatsApp is the built-in channel (via Baileys library, QR code pairing). Telegram, Discord, and Slack can be added via Claude Code skills — run /add-telegram, /add-discord, or /add-slack. Skills modify your fork's source code to add clean, dedicated channel support.

Security is a core design principle. Agents run in Linux containers (Apple Container on macOS, Docker on macOS/Linux) with filesystem isolation — not behind application-level permission checks. Secrets are passed via stdin JSON and never loaded into process.env. Mount allowlists with symlink escape detection. Containers run as non-root with read-only project mounts.

NanoClaw uses SQLite (better-sqlite3) for persistent storage — messages, groups, sessions, scheduled tasks, and router state. Each group also has its own CLAUDE.md file for per-group memory, mounted into its container. No external database needed.

Yes. NanoClaw runs on macOS and Linux — anywhere Node.js 20+ and Docker run. The lightweight TypeScript core works on low-resource devices including Raspberry Pi. Requires a container runtime (Apple Container or Docker) and an Anthropic API key.

Agent Swarms lets NanoClaw spin up teams of specialized agents that collaborate on complex tasks. NanoClaw is the first personal AI assistant to support agent swarms, built on Claude Code's agent-teams capability.

The project is open-source under MIT license at github.com/qwibitai/nanoclaw. Check the README for contributing guidelines. All contributions welcome.

Built-in tools are provided by Claude Code running inside containers: shell commands, file read/write, web search, web browsing (Chromium), and agent delegation (swarms). The host process provides scheduled tasks (cron), persistent SQLite memory, and WhatsApp messaging. Extend functionality by adding skills.

Yes. Each agent container has Chromium installed and agent-browser for browser automation. Claude Code can search the web and browse pages for real-time information inside the sandboxed container.

The official and only trusted repository is github.com/qwibitai/nanoclaw. The community Discord is discord.gg/VDdww8qS42.

macOS and Linux. Requires Node.js 20+ and a container runtime — Apple Container on macOS or Docker on macOS/Linux. Windows is not officially supported but may work via WSL2 with Docker.

NanoClaw uses Anthropic Claude exclusively via the Claude Agent SDK. Set ANTHROPIC_API_KEY in your .env file, or authenticate via Claude Code OAuth. There is no multi-provider support — the SDK handles model selection internally.

The README in the GitHub repository (github.com/qwibitai/nanoclaw) is the primary documentation source. The docs/ directory contains architecture specs, security model, and debug checklist. The codebase is small enough that you can ask Claude Code to walk you through it.

Official channels: Discord (discord.gg/VDdww8qS42) and GitHub (github.com/qwibitai/nanoclaw). Check the README for all community links.