NanoClaw

NanoClaw is a lightweight, open-source personal AI agent that runs on your own machine. It connects to messaging apps like WhatsApp, Telegram, Slack, Discord, and Signal, runs every agent session inside an isolated container (Docker or Apple Container), and is built around just 15 source files you can read in a single sitting. It's built on the Claude Agent SDK and designed for people who want to own and fully control their AI assistant.

OpenClaw is a monolithic framework with thousands of source files and dozens of dependencies. NanoClaw takes the opposite approach: ~15 source files, a single Node.js process, and real OS-level container isolation instead of application-level permission checks. If you want something you can fully audit, understand, and customize, NanoClaw is built for that.

Security is a core design principle. Every agent session runs inside an isolated Linux container with its own filesystem, IPC namespace, and process space. Agents can only access directories that are explicitly mounted, and there's no ambient access to your system. Because the entire codebase is around 3,900 lines of code across 15 files, you can realistically audit everything the software does. This combination of container-level isolation and a small, auditable codebase makes NanoClaw one of the most transparent and secure ways to run AI agents.

No. NanoClaw uses an AI-native setup process where Claude Code handles installation, configuration, and ongoing changes interactively. You run the /setup skill and follow the prompts. If you want to customize behavior later, you describe what you want in plain language and Claude Code modifies the codebase for you. That said, being comfortable with a terminal and running a few commands like git clone is expected.

NanoClaw itself is completely free and open source under the MIT license. However, it runs on the Claude Agent SDK, which requires a Claude API key or a Claude Code subscription. The cost depends on how much you use it. NanoClaw is designed to be lightweight in token usage, but the underlying AI usage is billed by Anthropic.

NanoClaw ships with WhatsApp support out of the box and can be extended to work with Telegram, email, and other channels through the skills system. Adding Telegram is as simple as running the /add-telegram skill, which walks you through the full setup. The architecture is designed so that messaging channels are modular, so you can run one or multiple channels side by side, each with their own configuration.

NanoClaw supports both Apple Container and Docker. On macOS, Apple Container provides fast, lightweight sandboxing optimized for Apple silicon, while Docker is available as an alternative. On Linux, Docker is used automatically. During the initial /setup process, you choose which runtime to use based on your platform and preferences.

Yes. NanoClaw runs on both macOS and Linux. On Linux, Docker is used as the container runtime and is configured automatically during setup. The system requirements are Node.js 20 or later, Claude Code, and a container runtime (Docker on Linux). The /setup skill handles all dependency checks and configuration for your platform.

NanoClaw uses an AI-native setup process. After cloning the repository and running Claude Code, you run the /setup skill, which guides you through installing dependencies, authenticating your messaging channels, configuring your container runtime, and starting the background services. There are no manual configuration files to edit. Claude Code handles the entire setup interactively. For ongoing changes, you can describe what you want and Claude Code will make the modifications directly in the codebase.

Most AI agent frameworks are designed for teams building products. They're large, complex, and require significant investment to understand. NanoClaw is designed for individuals who want a personal AI assistant that they fully own and control. It runs as a single Node.js process, uses real container isolation rather than application-level sandboxing, and is small enough to understand completely. It runs on the Claude Agent SDK, giving you direct access to Claude's capabilities without abstraction layers.

Yes. Nano Claw is fully open source under the MIT license. The entire codebase is available on GitHub, and contributions are welcome. The project encourages forking and customization. The philosophy is that your personal AI agent should be working software tailored to your exact needs, not a generic framework you configure.