What nanobot is
nanobot, from the HKUDS lab (University of Hong Kong), is an open-source, self-hosted personal AI agent framework written in Python. On October 12, 2026 the repo had about 49,000 stars on GitHub, an MIT license and commits from the previous day, per the GitHub API. The pitch: an assistant like the commercial ones, but running on your machine or server, with a small, readable core.
What it does
Per the README, nanobot:
- runs in a browser WebUI or the terminal;
- connects to Telegram, Discord, Slack, WeChat, email, Mattermost, Linear, Feishu, Teams and more;
- uses tools for files, shell, web search and fetch, MCP, cron, image generation and subagents;
- keeps session history and long-term memory (a feature called Dream);
- runs long-horizon goals and scheduled automations;
- exposes a Python SDK and an OpenAI-compatible API;
- works with OpenAI-compatible APIs, local LLMs and fallback models.
How to install
Requires Python 3.11+:
# with uv (recommended for stable)
uv tool install nanobot-ai
# or pip, inside a virtual environment
python -m pip install nanobot-ai
There's also a one-line installer (curl ... | sh). If you use it, read the script first (scripts/install.sh) or use --dry-run, which the README itself suggests to preview the plan.
Then:
nanobot webui
This opens http://127.0.0.1:8765. In Settings → Models, pick provider, credential and model. To keep it running after closing the terminal: nanobot gateway --background.
Strengths
- Data control: conversations and memory stay on your infrastructure.
- Model freedom: a local model for sensitive data, a cloud model for general tasks.
- Chat channels: handy for teams living in Slack, Teams or Telegram.
- Small, readable code: easier to audit than giant platforms.
Security cautions
An agent with shell, files and email access is powerful, and dangerous if misconfigured:
- Keep the WebUI on localhost. The README says first run binds to localhost only. Don't expose it without auth and TLS.
- Run it as an isolated user or container, with access only to the folders it needs.
- Watch for prompt injection: emails and web pages the agent reads can carry malicious instructions. Require human confirmation for sensitive actions.
- Store API keys in environment variables or a vault, never in repos.
- Review third-party MCP servers before connecting.
Who it's for
Developers and small technical teams wanting an assistant under their control, wired into company chat and local models. Without a technical team, a managed assistant may be safer and simpler.
Sources
- Repository: https://github.com/HKUDS/nanobot
- Project README and docs (checked Oct 12, 2026)
- GitHub API (stars and license, Oct 12, 2026)
Want to deploy an AI agent safely in your company? KGA IT offers a free consultation: kga-it.com.