neumann-vector

What problem does it solve?

Provide a clear, operational guide for storing embeddings, building HNSW indexes, and performing fast approximate nearest neighbor queries so developers can implement similarity search, RAG retrieval, and semantic caching without ad hoc tooling.

Core Features & Use Cases

• Vector storage model: Keyed embeddings stored in a dedicated vector engine with optional collection namespaces and sharded locking for concurrent writes.
• ANN search with HNSW: Explicit index build step for fast SIMILAR queries supporting COSINE, EUCLIDEAN, and DOT_PRODUCT metrics.
• Cross-engine patterns: Combine vector similarity with graph traversal and semantic caching to power RAG, agent memory, and similarity-aware graph operations.
• Performance & safety tips: Guidance on batch loading, index build timing, consistent dimensionality, and using collections for multi-tenant isolation.

Quick Start

Store embeddings, build the HNSW index, and run a SIMILAR query to retrieve top-k results for your application.