← Back to research
·4 min read·opensource

NanoBot

NanoBot is an ultra-lightweight Python personal AI assistant from Hong Kong University, delivering OpenClaw functionality in 4,000 lines of code with support for 9+ chat platforms.

Key takeaways

  • 4,000 lines of Python delivers core OpenClaw functionality — 99% smaller codebase
  • Most platform integrations of any lightweight alternative: Telegram, Discord, WhatsApp, Slack, Feishu, DingTalk, Email, QQ
  • MCP support and ClawHub skill integration for extensibility
  • pip install simplicity — running in under 2 minutes

FAQ

How does NanoBot compare to OpenClaw?

NanoBot delivers core functionality in ~4,000 lines of Python vs OpenClaw's 430,000+ lines of TypeScript. Trade-off: smaller feature set but much easier to audit and understand.

What platforms does NanoBot support?

Telegram, Discord, WhatsApp, Slack, Feishu, DingTalk, Email, QQ, and Mochat — more than any other lightweight alternative.

Who built NanoBot?

Hong Kong University of Science and Technology (HKUDS) research team.

Overview

NanoBot is an ultra-lightweight personal AI assistant from Hong Kong University, delivering OpenClaw's core functionality in approximately 4,000 lines of Python — 99% smaller than OpenClaw's 430,000+ line codebase.[1]

With 22,000+ GitHub stars, NanoBot has the largest community among OpenClaw alternatives. The project prioritizes simplicity, readability, and research-friendliness over feature completeness.

Key Stats

MetricValue
GitHub Stars22K+
LanguagePython
Core Lines~4,000
Installpip install nanobot-ai
First ReleaseFebruary 2, 2026
Latestv0.1.4 (Feb 17, 2026)

Installation

# Install from PyPI
pip install nanobot-ai

# Or with uv (faster)
uv tool install nanobot-ai

# Initialize
nanobot onboard

# Start chatting
nanobot agent

# Or run gateway for chat platforms
nanobot gateway

Platform Support

NanoBot supports more chat platforms than any other lightweight alternative:

PlatformAuth Method
TelegramBot token from @BotFather
DiscordBot token + Message Content intent
WhatsAppQR code scan
SlackBot token + App-Level token
FeishuApp ID + App Secret (WebSocket)
DingTalkApp Key + App Secret (Stream Mode)
EmailIMAP/SMTP credentials
QQApp ID + App Secret
MochatClaw token (auto-setup available)

Provider Support

ProviderNotes
OpenRouterRecommended for global access
AnthropicClaude direct
OpenAIGPT models
DeepSeekChinese provider
GroqFast inference + free Whisper
GeminiGoogle models
vLLMLocal model support
MiniMaxChinese provider
Moonshot/KimiChinese provider
CustomAny OpenAI-compatible endpoint

Features

  • Memory System — Built-in persistent memory (redesigned Feb 12)
  • MCP Support — Model Context Protocol integration (added Feb 14)
  • ClawHub Skills — Search and install public agent skills
  • Scheduled Tasks — Natural language task scheduling
  • Voice Transcription — Telegram voice via Groq Whisper (free)
  • Agent Social Network — Join Moltbook and ClawdChat communities

Architecture

User → Chat Platform → NanoBot Gateway → LLM Provider → Response
                           ↓
                    Memory System (SQLite)

Single Python process. No microservices. No message queues. Minimal dependencies.

Strengths

  • Simplicity — 4,000 lines you can read in a weekend
  • Platform Coverage — 9 chat integrations, more than any alternative
  • pip Install — Running in under 2 minutes
  • Research-Friendly — Clean, readable code for modification
  • Active Development — Multiple releases per week
  • MCP Support — Extensible via Model Context Protocol
  • Local Models — vLLM integration for self-hosted LLMs

Weaknesses / Considerations

  • Python Runtime — 100MB+ RAM overhead vs Go/Rust alternatives
  • Startup Time — 30+ seconds vs sub-second for compiled alternatives
  • Security — Application-level, not OS-level isolation
  • Feature Set — Smaller than full OpenClaw
  • Single Process — No multi-agent architecture

Who Built It

The HKUDS (Hong Kong University of Science and Technology Data Science) research team. The project is designed to be research-ready — clean code that's easy to understand, modify, and extend for academic purposes.

Ideal User

  • Researchers wanting readable AI agent code to study and modify
  • Multi-platform users needing support for obscure chat apps (Feishu, DingTalk, QQ)
  • Python developers who prefer pip-install simplicity
  • Teams wanting quick deployment without infrastructure complexity

Bottom Line

NanoBot is the "good enough" OpenClaw alternative for most users. The 99% reduction in code complexity makes it auditable, the pip-install process makes it accessible, and the 9+ platform integrations cover nearly any chat app.

The trade-off is Python's runtime overhead — 100MB+ RAM and 30+ second startups are fine for a VPS but problematic for edge deployment. For constrained hardware, look at PicoClaw (Go) or ZeroClaw (Rust). For maximum features, stick with OpenClaw.

For users who prioritize understanding their code over maximum capability, NanoBot hits the sweet spot.