JARVIS (HuggingGPT) is a collaborative system that uses an LLM as a controller to connect numerous expert AI models from Hugging Face to solve complicated AI tasks.
JARVIS is a system whose mission is to explore artificial general intelligence (AGI) and deliver cutting-edge research to the community. It introduces a collaborative system consisting of an LLM as the controller and numerous expert models as collaborative executors from the HuggingFace Hub. The workflow consists of four stages: Task Planning, Model Selection, Task Execution, and Response Generation. Jarvis can plan tasks, schedule Hugging Face models, and generate friendly responses based on user requests.
Exploring artificial general intelligence (AGI) and delivering cutting-edge research to the community; using an LLM as a controller to connect and orchestrate numerous expert AI models from Hugging Face to solve complicated AI tasks through task planning, model selection, task execution, and response generation.
JARVIS/HuggingGPT is described in its README as a multi-stage agentic system where an LLM controller autonomously performs task planning, model selection, task execution, and response generation. The system invokes and executes selected expert models without per-step human confirmation — a single user request triggers a full autonomous pipeline. Surface analysis confirms the code executes arbitrary code and shell commands, consistent with autonomous multi-step task execution. While the entrypoint code showed no explicit autonomy signals, the documented workflow and surface findings clearly describe an autonomous agentic loop, placing it at Tier 4. It does not modify its own code/prompts or spawn persistent new agents, so it does not reach Tier 5.