ictnlp/Auto-RAG

Implements autonomous iterative retrieval where LLMs dynamically decide when and what to retrieve through multi-turn dialogue, terminating retrieval once sufficient knowledge is available. Built on FlexRAG framework with vLLM for inference, supporting dense retrieval via FAISS indexing and HuggingFace encoder integration. Provides both browser-based GUI and programmatic interfaces for interactive RAG workflows with configurable metrics evaluation across multiple benchmarks.

235 stars. No commits in the last 6 months.