nanobot is an ultra-lightweight personal AI agent runtime released in February 2026 by the HKUDS (Hong Kong University Data Intelligence Lab). Similar to how GPT processes any text, nanobot is designed to enable any LLM to operate within a single, consistent agent loop. It implements the core agent pipeline — message reception, LLM decision-making, tool invocation, and context management — with minimal Python code, while also adopting a modular structure that allows for the addition of practical components such as WebUI, messenger integration, MCP (Model Context Protocol) hosting, long-term memory, and periodic scheduling. The limitation of existing agent frameworks is their complexity. Large frameworks like LangChain and AutoGen stack thousands of lines of abstraction layers, making it difficult for researchers to understand the internal workings or replace specific components. nanobot takes the opposite approach. It maintains the core logic of the agent loop at a readable level of a few thousand lines, while also enabling the switching between more than 30 LLM providers (OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Ollama, vLLM, LM Studio, AWS Bedrock, Azure OpenAI, etc.) using named presets and fallback routing. By connecting multiple MCP servers, external tools can be exposed to the agent, and the context can be automatically compressed and restored in long-term sessions using a token-based, two-stage memory system (Dream). Furthermore, it supports natural language-based cron scheduling, sandbox code execution, and background execution of sub-agents, allowing for the construction of agents that continuously track goals beyond simple chatbots. From the perspective of a biotechnology researcher, the value of nanobot lies in its ability to "quickly build a personal AI assistant without infrastructure dependencies." For example, an agent can be created to monitor research papers, connect the PubMed API as an MCP tool, collect new papers daily via cron, and complete the pipeline with a few dozen lines of code to deliver summaries to Telegram or Slack. In experimental data preprocessing workflows, a local LLM (such as Ollama) can be connected to the agent to automate tasks such as CSV cleaning, statistical summarization, and visualization script generation through natural language instructions, without sending sensitive data to external APIs. The WebUI is embedded within the Python wheel, so a browser-based management screen can be opened with `nanobot run` without separate building, and Docker container deployment is also supported, making it suitable for operating a research support agent that runs continuously on a lab server.