NanoClaw is a secure, lightweight, and open-source personal AI agent designed for individual users who prioritize data ownership and system control. Unlike monolithic AI agent frameworks that require massive infrastructure and complex configuration, NanoClaw is built on a minimal architecture consisting of approximately 15 source files and a single Node.js process. This design allows users to realistically audit, understand, and customize the entire codebase to fit their specific requirements, moving away from complex, opaque systems towards bespoke, maintainable software.

Functionality is managed through a modular, AI-native approach where the system ships as a minimal trunk and relies on self-registering extensions called "skills." Users can expand the agent's capabilities—such as adding support for messaging channels like WhatsApp, Telegram, Slack, and Discord, or integrating tools like Gmail, GitHub, or web browsing—on demand. Each channel or skill is installed via simple commands that dynamically inject the necessary modules directly into the user's fork, ensuring the software remains lean and free of unnecessary bloat.

Some of the key features are:

• Container Isolation: Agents run inside isolated Linux containers (or native Apple Containers on macOS), ensuring they can only access explicitly mounted host directories.
• OneCLI Agent Vault: Agents never handle raw API keys; instead, credentials are injected at request time through a secure vault that enforces per-agent policies and rate limits.
• AI-Native Setup: A scripted onboarding process guides users through installation, dependency management, and service startup, with automatic recovery and diagnosis via Claude Code if issues arise.
• Modular Channel Architecture: Support for diverse communication platforms including WhatsApp, Telegram, Discord, Slack, Microsoft Teams, and email is enabled via on-demand skill installation.
• Per-Agent Workspaces: Every agent group operates in its own container with a unique memory, CLAUDE.md workspace, and isolated process space to prevent data cross-contamination.
• Self-Documenting Codebase: With a tiny footprint of ~3,900 lines of code, the system is designed to be fully transparent, encouraging users to leverage AI assistance to walk through and modify the implementation.

Operationally, NanoClaw functions by routing inbound messages through an entity-based model that connects users, messaging groups, and specific agent sessions. All inbound traffic is processed via an SQLite-backed queue that orchestrates container lifecycles, ensuring no cross-mount contention or insecure IPC mechanisms are used. Because the architecture relies on local container execution rather than microservices, setup is straightforward, typically requiring only Node.js 20+, pnpm 10+, and Docker or a native runtime like Apple Container.

Some common use cases include:

• Personal AI Assistant: Automating daily briefings, managing schedules, and drafting emails via a private, containerized interface accessible through familiar chat apps.
• Collaborative Research: Using agents to crawl websites, scrape content, and compile findings while keeping the execution environment sandboxed and private.
• Development Workflow Automation: Integrating with platforms like GitHub or Linear to track issues, comment on pull requests, and deploy code directly from a chat interface.
• Cross-Platform Integration: Consolidating interactions from various messaging apps into a single, unified agent memory group for consistent AI performance.