← All agents
Nanobot logo

Nanobot

The ultra-lightweight personal AI agent — 99% less code, 100% of the essentials.

21k starsPython 3.11+by HKUDS

What is Nanobot?

Nanobot is an ultra-lightweight personal AI agent from the HKUDS research group. It was inspired by OpenClaw but stripped to the bone — the same core agent loop in 99% fewer lines of code. The point is readability: you can sit down and understand the whole thing end-to-end in an afternoon.

Its standout feature is a two-stage memory system: live conversation history is kept separate from consolidated long-term knowledge, with a background consolidation pass backed by git-versioned storage so every memory state is recoverable. MIT licensed.

Highlights

What you get when you run Nanobot on LaunchPod.

Tiny, hackable codebaseA personal AI agent you can actually read. Minimal dependencies, clean Python, easy to modify for your own research.
Two-stage memorySeparates live conversation from consolidated long-term knowledge. A background pass distils durable state.
Git-versioned memoryEvery consolidation is committed to git. You can roll back, diff, or audit your agent’s memory like any other repo.
Fast startup, low footprintNo heavy runtime, no Docker overhead. Runs cleanly on LaunchPod’s 4 GB shared-nothing VM.
Built for long-running agentsStability over features. The project optimises for stable, long-lived agent processes you’d actually leave running for weeks.
Actively maintainedRegular channel and provider additions, reliability fixes, and production-oriented improvements from HKUDS.

How it works on LaunchPod

LaunchPod installs nanobot in a Python 3.11 virtual environment on your dedicated VM, runs it as a systemd service, and exposes it behind a Cloudflare tunnel. You configure it the first time you open the browser terminal — no SSH client needed.

Resources

Nanobot is open source. Here's everything you need to dig deeper into the project.

Ready to run Nanobot?

First month free. $9/mo after that. 1,000 credits included.