nanobot — OpenClaw on a minimal scale

nanobot — OpenClaw on a minimal scale

OpenClaw — 430,000 lines of code, ~1 GB of RAM, Node.js. Not everyone has a server for this.

The guys from the University of Hong Kong threw out 99% of the code and kept the core: 3,700 lines in Python, ~100 MB RAM.

What it can do:
→ AI agent in Telegram (and WhatsApp, Discord, Slack)
→ Web search via Brave API
→ Cron tasks on schedule
→ Memory between sessions
→ Sub-agents for background tasks
→ 11+ LLM providers (Claude, GPT, Gemini, DeepSeek, local models via vLLM)
What it lacks: browser, 5700 skills, voice mode, smart home, Canvas. All of this remained in OpenClaw.

Installation — two minutes:
pip install nanobot-ai
nanobot onboard
I entered the OpenRouter API key, Telegram bot token — done.
100 MB RAM means this thing will fit on a router with OpenWrt or on the cheapest VPS for $3/month. OpenClaw cannot do that.

Who it’s for: if you need a simple AI bot in Telegram with task scheduling and don’t need all the power of OpenClaw — nanobot covers 80% of scenarios with 1% of resources.
github.com/HKUDS/nanobot