explaingit

qwibitai/nanoclaw

📈 Trending★ 28,642TypeScriptAudience · developerComplexity · 3/5ActiveLicenseSetup · moderate

TLDR

A lightweight AI assistant that runs in isolated containers and connects to messaging apps like WhatsApp, Telegram, and Slack. Small, auditable codebase you can read and modify yourself.

Mindmap

mindmap
  root((NanoClaw))
    What it does
      AI agent in containers
      Messaging app integration
      Sandboxed execution
    How to use
      One-line setup script
      Trigger word activation
      Web search and files
    Tech stack
      TypeScript
      Docker containers
      Claude AI models
    Use cases
      Personal assistant
      Team automation
      Task execution
    Design philosophy
      Auditable codebase
      Code-based customization
      Modular skills

Things people build with this

USE CASE 1

Set up a personal AI assistant that responds in your Slack or Discord channels when you mention it by name.

USE CASE 2

Run automated tasks and web searches inside a sandboxed container so the AI cannot access your main system.

USE CASE 3

Customize your agent's behavior by editing the TypeScript source code directly with Claude's help.

Tech stack

TypeScriptDockerClaude APINode.js

Getting it running

Difficulty · moderate Time to first run · 30min

Requires Claude API key and messaging app credentials (WhatsApp/Telegram/Slack) to connect.

Use freely for any purpose including commercial, as long as you keep the copyright notice.

In plain English

NanoClaw is a self-hosted AI assistant that lets you talk to Claude through messaging apps. The README presents it as a lightweight alternative to OpenClaw, written to be small enough that one person can read and modify the whole codebase. The author's concern with OpenClaw is that it has roughly half a million lines of code and relies on application-level allowlists rather than OS-level isolation; NanoClaw instead runs each agent inside its own Linux container, so the agent can only see files you explicitly mount. You install it by cloning the repository and running a setup script that installs Node, pnpm, and Docker if needed, registers your Anthropic credential with a tool called OneCLI, builds the agent container, and pairs your first channel. The README lists supported channels including WhatsApp, Telegram, Discord, Slack, Microsoft Teams, iMessage, Matrix, Google Chat, Webex, Linear, GitHub, WeChat, and email through Resend, installed on demand via "/add-channel" skills. Each agent has its own workspace with its own CLAUDE.md, memory, and container. The README also describes scheduled recurring tasks, web search and fetch, and credential routing through OneCLI's Agent Vault so the agent never holds raw API keys. NanoClaw is built on Anthropic's official Claude Agent SDK and uses Claude Code as its harness, with drop-in support for other providers via additional skills. Customisation is done by telling Claude Code in plain English what to change rather than editing config files. Someone would use this if they want a personalised, self-hosted assistant reachable from their normal chat apps with strong sandboxing. The primary language is TypeScript.

Copy-paste prompts

Prompt 1
Help me set up NanoClaw to connect to my Slack workspace and respond when I type @assistant in channels.
Prompt 2
Show me how to add a new skill to NanoClaw that lets my AI agent query a custom API.
Prompt 3
I want to modify NanoClaw's prompt so Claude acts like a project manager. What files do I edit?
Prompt 4
How do I configure NanoClaw to run on my Docker setup and connect it to WhatsApp?
Open on GitHub → Explain another repo

Generated 2026-05-21 · Model: sonnet-4-6 · Verify against the repo before relying on details.