gomate-community/TrustRAG

Implements a modular RAG pipeline with configurable components for document parsing, retrieval, reranking, and generation, supporting hybrid retrieval combining BM25 and dense embeddings. Features DeepSearch—a recursive query decomposition framework with token budgeting and action-based reasoning (search, reflect, answer, read, coding)—alongside multimodal QA, vector database integration (Milvus, Qdrant), and web search capabilities. Targets Python 3.10+ and integrates with LLMs like GLM-4 and OpenAI via configurable APIs.

1,233 stars and 31 monthly downloads. Available on PyPI.