RAGFlow 0.27.1 pushes metadata filters down to the metadata index and exposes rerank candidate counts

GitHub·medium signalXBlueskyLinkedInCopy link

The 2026-08-28 release moves metadata filtering into the metadata index rather than applying it post-retrieval, and opens rerank_candidates_count, knn top_k and num_candidates on the Retrieval API so the recall/precision tradeoff is tunable per call instead of fixed. It adds an Azure DevOps data-source connector and You.com and Serply as web search providers, plus AWS Bedrock API key authentication. Bug fixes include a QA Excel parser dropping rows and wiki compilation accepting LLM error responses as valid output.