JARVIS is a Microsoft research system that connects large language models with the machine learning community to explore AGI. It enables task planning and decomposition, dynamically selects appropriate ML models for subtasks, and integrates various AI tools including EasyTool for concise tool instruction and TaskBench for evaluating LLM task automation capability. The system supports multiple LLM backends including GPT-4 and has been integrated into LangChain as an autonomous agent implementation.