Nanobot
Lightweight, open-source AI agent for your tools, chats, and workflows.
Last verified:
What is Nanobot?
Nanobot is an ultra-lightweight, self-hosted personal AI agent framework developed by the Hong Kong University Data Science Laboratory (HKUDS). It implements over 90% of OpenClaw's core capabilities in just approximately 4,000 lines of code - a 99% reduction from OpenClaw's 430,000+ lines. The tool helps users quickly deploy large language models and build 24/7 personal AI assistants for workspace automation, multi-platform messaging, and long-running workflows.
Key features include native MCP (Model Context Protocol) framework support with built-in MCP-UI integration for rendering interactive React components in chat clients. It supports 12+ messaging platforms including Telegram, WhatsApp, Discord, Slack, DingTalk, Feishu, QQ, and Email. Nanobot offers multi-model compatibility through LiteLLM with OpenAI, Claude, Gemini, DeepSeek, vLLM, and Groq. Built-in tools handle file operations, shell commands, web search, and web scraping, while the sub-agent system enables background task processing without blocking conversations.
Nanobot is designed for developers, researchers, and tech enthusiasts who want to understand AI agent internals without wading through massive codebases. It's ideal for personal knowledge management, development assistance, office automation, remote server operations, and deploying round-the-clock assistant bots. The pure Python implementation with pip install ready makes it accessible for low-resource deployments on Raspberry Pi and similar hardware.
The agent features long-term memory that automatically saves memories and daily notes across conversations, scheduled tasks via Cron expressions, and SKILL.md documentation for easy functionality extension. It can be deployed anywhere as a fully-featured MCP host and embedded into any application or website, with both local models (via vLLM) and online API access supported.
As an open-source MIT-licensed project in alpha status, Nanobot serves as an excellent educational tool for learning AI agent principles while providing production-ready functionality for personal automation needs.
Nanobot pricing
Pricing model: Freemium
Nanobot is completely free and open-source under the MIT license. No paid tiers or subscription plans exist. The tool can be installed via pip or Homebrew (brew install nanobot-ai/tap/nanobot) at no cost. All features including MCP support, multi-platform integration, multi-model compatibility, and the web UI are included without payment. Users only pay for external LLM API costs if using OpenAI, Anthropic, or other paid model providers - local models via vLLM are free.
Nanobot pros
- Ultra-lightweight at only ~4,000 lines of code - 99% smaller than OpenClaw
- Native MCP framework support built from the ground up
- First-class MCP-UI integration for interactive React components in chat
- Supports 12+ messaging platforms including WhatsApp, Telegram, Discord, Slack
- Multi-model support via LiteLLM with OpenAI, Claude, Gemini, DeepSeek, Groq
- Pure Python implementation with simple pip install - no complex compilation
- Raspberry Pi-friendly and optimized for low-resource deployments
- Long-term memory automatically saves across conversations
- Sub-agent system for background task processing without blocking
- Scheduled tasks support via Cron expressions
- Built-in tools for file operations, shell commands, web search, scraping
- Easy skill extension through SKILL.md documentation
- Flexible deployment as standalone MCP host or embedded in applications
- Clean modular code structure ideal for learning AI agent principles
- Open-source MIT license with active development community
Nanobot cons
- Still in alpha status with breaking changes expected
- Moving away from original design - architecture may shift significantly
- Evolving APIs that may not be stable
- Limited model provider support compared to commercial alternatives
- Production-ready UI still under development
- Multi-agent support needs improvement
- Authentication and security features need expansion
- Documentation available in limited languages despite 6+ language claim
Frequently asked questions about Nanobot
What is Nanobot and what does it do?
Nanobot is an ultra-lightweight personal AI agent framework developed by HKUDS (Hong Kong University Data Science Laboratory). It runs approximately 4,000 lines of code and implements over 90% of OpenClaw's core capabilities. Nanobot reads your workspace, edits files, runs commands, uses tools and APIs, and communicates through terminal or messaging platforms. It serves as a 24/7 personal AI assistant for automation, knowledge management, and workflow tasks.
How is Nanobot different from OpenClaw?
Nanobot reduces the codebase from OpenClaw's 430,000+ lines to approximately 4,000 lines - a 99% reduction. It addresses OpenClaw's pain points including massive codebase, steep learning curve, and complex deployment. While implementing over 90% of OpenClaw's core capabilities, Nanobot uses pure Python with clear modular structure, making it easy to understand, learn, and extend in resource-constrained environments.
What messaging platforms does Nanobot support?
Nanobot supports 12+ messaging platforms including Telegram, WhatsApp, Discord, Slack, DingTalk, Feishu, QQ, Email, and other IM platforms. The bot can be deployed on servers for round-the-clock message response across all these channels simultaneously.
Which LLM models can I use with Nanobot?
Nanobot supports multiple LLM providers through LiteLLM integration, including OpenAI (gpt-4, etc.), Anthropic (claude-3 series), Gemini, DeepSeek, vLLM for local models, and Groq. You set the corresponding API key (OPENAI_API_KEY or ANTHROPIC_API_KEY) and Nanobot automatically selects the correct provider based on the model specified.
How do I install Nanobot?
Nanobot can be installed via Homebrew on macOS: brew install nanobot-ai/tap/nanobot. This gives you the nanobot CLI for running and managing your MCP host. For Python installation, create a virtual environment and use pip install -e . after cloning from GitHub. The tool requires Python version >=3.9.
What is an MCP host and why does Nanobot need one?
An MCP host is the service that combines MCP servers with an LLM and context to present an agent experience. While applications like VSCode, Claude, Cursor, and ChatGuilt include MCP hosts, Nanobot is designed as a standalone, open-source MCP host that can be easily deployed or integrated. You can create your own dedicated MCP and MCP-UI powered chatbot using Nanobot as the host.
Can I run Nanobot on Raspberry Pi or low-resource hardware?
Yes, Nanobot is specifically optimized for low-resource deployments on Raspberry Pi and similar hardware. The ultra-lightweight ~4,000 line codebase and pure Python implementation make it ideal for resource-constrained environments where larger frameworks would struggle.
How does Nanobot handle long-term memory?
Nanobot automatically saves memories and daily notes across different conversations, providing long-term memory functionality. This allows the agent to maintain context and learn from previous interactions without requiring manual memory management.
Can I embed Nanobot into my existing application or website?
Yes, Nanobot can be deployed anywhere as a fully-featured MCP host and embedded into any application or website. The roadmap includes frontend integrations for Slack, SMS, email, and embedded web agents, with easy embedding into existing apps as a development goal.
Is Nanobot suitable for production use?
Nanobot is currently in alpha status with active development and rapid changes expected. The project is moving away from its original design, so expect significant breaking changes, architectural shifts, and evolving APIs. While it provides production-ready functionality for personal automation, the production-ready UI and robust multi-agent support are still in the roadmap.