bowang-lab/MedRAX

Combines specialized CXR analysis tools (classification, segmentation, grounding, report generation) through a LangChain/LangGraph-based agentic framework powered by GPT-4o vision. Includes ChestAgentBench, a 2,500-query benchmark across 7 diagnostic categories, enabling evaluation of multi-step medical reasoning without model retraining. Supports local LLM backends (Ollama, LM Studio) and OpenAI-compatible providers with selective tool initialization and 8-bit quantization for resource-constrained deployments.

1,108 stars.