Nanobot
An ultra-lightweight Python AI assistant from the Data Intelligence Lab at the University of Hong Kong (HKUDS). The entire core fits in roughly 4,000 lines of code—orders of magnitude smaller than OpenClaw—making it easy to read, audit, and modify.
Key Facts
- Language: Python
- Footprint: ~45MB, ~0.8s cold start
- Channels: Telegram, Discord, WhatsApp, Slack, Email, Feishu, DingTalk, QQ, Mochat
- LLM providers: 11+ including OpenRouter, Anthropic, OpenAI, DeepSeek, Google Gemini
- Tooling: File operations, shell execution, web access; MCP support (v0.1.4, Feb 2026)
- Memory: Long-term and short-term memory, cron-based scheduling
- License: MIT
Best For
Developers who want the smallest possible trustworthy codebase to build from, or teams that are MCP-first and want clean tool plugin support without the operational weight of a large runtime. Researchers exploring agent design also benefit from the readable architecture.
Note: Sandboxing is not enabled by default. The restrictToWorkspace flag exists but requires explicit configuration. The plugin ecosystem is thinner than OpenClaw's ClawHub. Not designed for enterprise deployment.
Compared to OpenClaw
Nanobot delivers similar core capabilities—multi-channel chat, tools, memory, scheduling—in a fraction of the code. You trade OpenClaw's 700+ skills marketplace and mature ecosystem for transparency and simplicity. If you need to understand or customize every line, Nanobot is a strong fit.