Overview

NanoClaw is a lightweight, open-source personal AI assistant designed to run securely on your own machine. Built as a direct response to the complexity and security concerns of larger platforms like OpenClaw, NanoClaw keeps its entire core to just ~15 source files (~3,900 lines of TypeScript) and fewer than 10 dependencies. Every AI agent session runs inside a true OS-level container (Docker on Linux/macOS or Apple Container on macOS) with strict filesystem isolation — nothing is shared unless you explicitly mount it.

It connects directly to messaging apps you already use (WhatsApp out-of-the-box, plus Telegram, Slack, Discord, Gmail and more via skills), maintains per-group memory, supports agent swarms for complex tasks, and includes built-in scheduled jobs (cron-style morning briefings, weekly reports, etc.). Setup and customization are completely AI-native: Claude Code reads the tiny codebase and handles everything.

Key Features

True Container Isolation: Each chat/group runs in its own sandboxed container with isolated filesystem, IPC namespace, and Claude session. No shared memory or application-level permission checks.
Multi-Channel Messaging: Native WhatsApp support; add Telegram, Slack, Discord, Signal or Gmail with simple /add-telegram-style skills.
Agent Swarms: Spin up teams of specialized agents that collaborate inside the same conversation — the first personal AI assistant with this capability.
Per-Group Memory & Context: Every group has its own CLAUDE.md memory file and isolated storage.
Scheduled Tasks: Recurring or one-off jobs that run Claude and can message you results.
Web Access & Tools: Built-in search, fetch, and browser capabilities via the Claude Agent SDK.
AI-Native Customization: No config files or dashboards — describe what you want and let Claude Code modify the codebase directly.
Single Process Simplicity: One Node.js orchestrator polls messages, manages queues, and spawns containers.

How It Works

Messages arrive via WhatsApp/Telegram/etc. → stored in SQLite.
Per-group queue routes the message to a fresh container.
Claude Agent SDK runs inside the isolated container with only the group's files mounted.
Output streams back; IPC happens safely via filesystem.
Concurrency is limited (default 3 containers) with automatic retries.

The entire architecture fits in your head in under 10 minutes — perfect for auditing and forking.

Installation & Quick Start

git clone https://github.com/qwibitai/nanoclaw.git
cd nanoclaw
claude

Inside the Claude Code prompt, simply type /setup. Claude Code will:

Install dependencies
Guide you through messaging app authentication
Configure your container runtime
Start background services

No manual YAML or config editing required. Runs on macOS and Linux (Node.js 20+ required; Claude API key or local model endpoint needed).

Security Model

NanoClaw's security is enforced at the operating-system level rather than inside application code:

Agents cannot see your full filesystem — only explicitly mounted directories.
Each group is sandboxed in its own container with separate process space and IPC namespace.
The tiny codebase (~3,900 LOC) makes full audits realistic.
Full details in docs/SECURITY.md (included in repo).

Compared to OpenClaw's single-process shared-memory model, NanoClaw eliminates entire classes of risk.

Skills & Extensibility

Instead of bloating the core, functionality is added through reusable Claude Code skills (e.g. /add-gmail, /add-signal). Community members contribute skills that transform your personal fork — resulting in clean, purpose-built code with zero unused modules.

Use Cases

Personal sales pipeline management via WhatsApp
Daily morning briefings and research summaries
Automated weekly reports from multiple data sources
Secure multi-agent collaboration on complex projects
Self-hosted AI assistant that respects your privacy boundaries

Tech Stack & Requirements

Language: TypeScript / Node.js (single process)
AI: Anthropic Claude Agent SDK (Claude API key required)
Containers: Docker or Apple Container
Database: SQLite
License: MIT
GitHub: qwibitai/nanoclaw (21k+ stars)

NanoClaw is intentionally minimal — built for users who want full ownership, transparency, and the ability to customize every detail with the help of the world's best coding model.