Skip to content
← Tools / Skills
ToolHKUDS (University of Hong Kong)/10 trustMetrics-checked

NanoBot

Ultra-lightweight personal AI agent in ~4,000 lines of Python — 99% smaller than OpenClaw, MIT-licensed.

Visit official site →

About

NanoBot is an open-source personal AI agent framework from the HKUDS data-intelligence lab, released in February 2026 as a deliberate counterpoint to OpenClaw's 430,000-line codebase: the entire agent runtime fits in roughly 4,000 readable lines of Python. It has become one of the fastest-growing agent projects of 2026 with over 49,000 GitHub stars and commits landing as recently as this week.

Small does not mean toy. One nanobot.yaml wires the same agent into Telegram, Discord, WhatsApp, Slack, email, and more with shared memory; it supports 11+ LLM providers (OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, local vLLM, and others), MCP servers, cron-scheduled tasks, subagents, shell execution, and voice input. A browser WebUI, a terminal client, a Python SDK, and an OpenAI-compatible HTTP API round it out, and it runs anywhere from a Raspberry Pi to a Kubernetes cluster on about 100MB of RAM.

What it can do

  • Full agent loop (reasoning, tools, memory, scheduling) in ~4,000 lines of readable Python
  • 8+ chat channels (Telegram, Discord, WhatsApp, Slack, Feishu, QQ, email) sharing one session state
  • 11+ LLM providers plus local models via Ollama/vLLM/LM Studio
  • MCP servers, Agent Plugins, CLI apps, and reusable Skills
  • Cron-scheduled and recurring tasks; long-term memory across restarts
  • Browser WebUI, terminal client, Python SDK, and OpenAI-compatible HTTP API

Setup

  1. Install Python 3.10+.
  2. Run pip install nanobot-ai (or uv tool install nanobot-ai).
  3. Write a nanobot.yaml with your model provider key and channel bot tokens.
  4. Run nanobot start — the agent is now reachable on every configured channel with shared memory and MCP tools.

Prerequisites

Python 3.10+; at least one LLM provider API key (or a local OpenAI-compatible server); bot tokens for whichever chat channels you want.

Use cases

agentsautomationchatopen-source

Works with

Trust score — 9/10

/10 trustMetrics-checked

Scored 9/10 (metrics tier, metrics-v1). Evidence: 49,062 GitHub stars (verified via GitHub API), MIT license, commits as of Oct 15 2026, independent tutorials and reviews corroborating the feature claims. Strengths: genuinely open, tiny auditable codebase, multi-channel and multi-provider support verified across sources. Docked for: (1) academic-lab origin with a bus factor concentrated on one maintainer; (2) the '99% smaller than OpenClaw' framing is the project's own marketing — directionally true but self-reported. License: MIT.

★ 49,062 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 49,062 GitHub stars (log-scaled vs catalog peers, verified via GitHub API)
  • · last push Oct 15 2026; MIT-licensed; ~4,000-line Python codebase is genuinely auditable
  • · 8+ chat channels sharing session state, 11+ LLM providers, MCP servers, cron scheduling — corroborated by independent tutorials and reviews
  • · manual review: academic-lab origin, bus factor concentrated on one maintainer; the '99% smaller than OpenClaw' framing is the project's own marketing

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.