nanocoai

nanoclaw: Run Claude Agents in Isolated Containers

NanoClaw is a lightweight, customizable AI assistant built on Anthropic’s Claude Agent SDK. It runs agents in isolated containers and connects them to messaging apps on macOS, Linux, or Windows via WSL2.

ActiveTypeScriptMITAI AgentsSecurityDocker
Stars
31.1k
Forks
12.7k
Stars, last 30 days
+143
Last commit
6 days ago
Contributors
100+

Our take

Activeregular commits and releases

Health score

70/100

How it is scored
Activity
76
Community
79
Issues
66
Pull requests
57

+143 stars in the last 30 days

  • Merges more pull requests than 80% of the projects we track
  • More contributors than 78% of the projects we track

NanoClaw is an actively maintained, customizable Claude assistant for users willing to run and adapt a container-based system.

Good fit if

  • You want a personal assistant built on Anthropic’s Claude Agent SDK, with agents isolated in containers.
  • You are comfortable with Node.js, Docker-based setup, and customizing a codebase to suit your needs.
  • You value a project with regular recent commits and releases, and contributions from many people rather than dependence on one maintainer.

Look elsewhere if

  • You need a ready-to-use managed assistant or a dashboard for monitoring and debugging, which NanoClaw does not provide.
  • You want to configure a generic product without changing code, or do not want to run Docker containers.
  • You need a built-in channel or provider that is not available for your fork through the project’s skills and channel branch.
All health signals
Last commit2026-10-26 (6 days ago)
Commits, last 90 days100+
Releases, last 12 months13 (latest v2026.10.1-rc.2, 2026-10-22, pre-release)
Contributors100+ (top contributor: 50% of commits)
Issues closed, last 90 days4+ (typically closed in 1 day)
Pull requests merged, last 90 days66 (typically merged in 1 day)
Project age9 months

Checked on 2026-11-02 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 9, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

31,137

Forks

12,639

Change since Nov 21, 2026: +5 stars·+2 forks

Overview

NanoClaw is a lightweight AI assistant built on Anthropic’s Claude Agent SDK. It runs agents in separate Linux containers to limit filesystem access, addressing the risk of giving an assistant broad access to a host machine. It supports macOS and Linux, with Windows via WSL2.

The project is designed for individual users who want to connect assistants to messaging apps, give each agent its own memory and workspace, and customize behavior by changing code. Its setup flow uses Docker and can pair agents with channels such as Slack, Telegram, Discord, WhatsApp, or a local CLI.

Key Features

  • Runs each agent group in its own container, with only explicitly mounted files available to it.
  • Uses Anthropic’s Claude Agent SDK as its built-in agent runtime.
  • Routes messages between supported messaging channels and agents, with channel adapters installed on demand.
  • Provides per-agent memory, workspace instructions, and configurable channel-to-agent isolation.
  • Schedules recurring tasks and can use script gates to avoid waking an agent when no work is due.
  • Routes outbound requests through a credential gateway, keeping raw API keys out of agent containers.
  • Supports agent templates and code-based customization of the installation.

Use Cases

  • A privacy-conscious individual can run a personal assistant in a container and grant it access only to selected folders.
  • A team or community administrator can let members interact with agents through messaging channels such as Slack or Discord.
  • A user can schedule recurring briefings or reminders and receive them in a messaging app.
  • A developer can fork NanoClaw and modify the assistant’s behavior directly instead of managing a large configuration system.

What you need

Detected in the repository

  • Node.js >=22 (from package.json)
  • Automated checks on GitHub Actions

License in plain words

MITpermissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the copyright and license notice
  • Share your changes: no

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

git clone https://github.com/nanocoai/nanoclaw.git nanoclaw
cd nanoclaw
bash nanoclaw.sh

See the README for setup requirements, channel pairing, and customization details.

Alternatives

  • mercury-agent: Mercury is a self-hosted assistant for CLI, Telegram, and web, while NanoClaw focuses on messaging integrations and isolated agent containers.
ProjectLanguageLicenseStarsStatus
nanoclawTypeScriptMIT31.1kActive
mercury-agentTypeScriptMIT3.2kNew

Considerations

NanoClaw is active, with regular recent commits and releases, and pull requests are generally merged quickly. Its contributor base is broad rather than concentrated around one person, but the project’s rapid development means behavior and setup may change; check the changelog before upgrading. The measured health data also shows relatively few issues closed recently, so do not assume every reported problem will be resolved promptly.

The README describes Docker as the container runtime and lists Node.js 22 or later, pnpm, and Claude Code for setup recovery and customization. Review the security model and mounted directories before connecting an agent to sensitive files. The README says NanoClaw does not include a monitoring dashboard or debugging UI, and expects users to use Claude Code for troubleshooting. Channel adapters and alternative providers may require installing skills rather than being included in a default setup.

Found this useful?

Share it with someone who would like nanoclaw.

Comparisons