reasonkit/reasonkit-mem

Based on the README, here's a technical summary: **Combines Qdrant dense vectors with Tantivy BM25 sparse indexing for hybrid search fusion, supporting both local BGE-M3 and remote OpenAI embeddings.** Implements RAPTOR hierarchical tree indexing for recursive abstractive processing across large document collections, enabling multi-level question answering. Built as a Rust async library with embedded mode (file-based fallback) or remote Qdrant server, integrating reranking, document chunking, and knowledge base abstractions into ReasonKit's reasoning pipeline.