Skip to content

Summary

Open-source, self-hosted personal AI agent you run yourself — browser UI, terminal or your chat apps, with memory, MCP tools and scheduled automations.

Screenshots

Description

What it is

nanobot is a self-hosted personal AI agent runtime written in Python. Instead of signing into somebody's assistant, you run the agent on your own machine or server and reach it however you like: a bundled browser WebUI, the terminal, or the chat app you already live in — Telegram, Discord, Slack, WeChat, Feishu, Microsoft Teams, Mattermost or plain email.

It comes from HKUDS, the data science research group at the University of Hong Kong, and is MIT licensed.

What it actually does

The agent has a working toolbelt rather than just a chat box: file access, a shell, web search and fetch, image generation, cron scheduling, subagent delegation, and MCP so it can pick up any server from the wider ecosystem. Long-term memory is handled by a subsystem the project calls Dream, which keeps context alive across sessions rather than resetting each conversation.

That combination is what makes it a runtime rather than a chatbot. Long-horizon goals and scheduled automations keep running between your conversations, so you can hand it something recurring and check the result later. Model routing with fallbacks means it can drive OpenAI-compatible APIs or a local LLM, and switch when one is unavailable.

Who it suits

The pitch is ownership. Everything is inspectable and self-hosted, the core is deliberately small and readable, and there is no platform between you and the model provider — you bring your own API keys or point it at a local model. If you want a persistent assistant with access to your files and shell, running it yourself is a very different privacy posture from a hosted product, and nanobot is built around that assumption.

It is equally a base to build on: there is a Python SDK and an OpenAI-compatible API endpoint, so nanobot can act as an agent gateway that your own applications call.

Getting started

Free and open source under MIT. Requires Python 3.11 or newer; the package is nanobot-ai on PyPI and installs via pip, uv, Docker, or one-command setup scripts for macOS, Linux and Windows. Version 0.3.0, "The Agency Release", landed in July 2026. Documentation at nanobot.wiki is published in ten languages, and the project maintains a guide for installing with no terminal or configuration background — worth knowing, because self-hosting is otherwise the main barrier here. Running costs are whatever your chosen model provider charges; nanobot itself charges nothing.

Reviews

No reviews yet — sign in to write the first one

Similar App Suggestions

App: Langflow

IBM (DataStax)

Open-source, MIT-licensed visual builder for AI agents, RAG apps and MCP servers, maintained by IBM's DataStax and extensible in Python.

Automation & WorkflowsFree

Open-source platform for running a standing team of AI agents — you hire, schedule and review them, and they keep working while you are offline.

Business & ProductivityFreemium

Secure cloud sandboxes where AI agents can run untrusted code safely, starting in milliseconds.

Coding & DevelopmentFreemium

App: seendiff

Abhishek Iyer

Featured

A local, keyboard-driven diff viewer built for the huge changes coding agents produce - and it remembers what you have already seen.

Coding & DevelopmentFree

App: oMLX

Jun Kim

Featured

Menu-bar LLM inference server for Apple Silicon, with continuous batching and tiered KV caching that keeps local models fast enough for real coding work.

Coding & DevelopmentFree