AgentDB Vector Search

Semantic vector search for a local knowledge base is implemented using AgentDB with HNSW indexing, enabling sub-millisecond retrieval across thousands of documents. You initialize a database and run queries using an embedding model to produce vectors.

Yes, you can build retrieval augmented generation (RAG) pipelines using AgentDB for intelligent document retrieval. It provides RAG-ready integration by storing document embeddings and retrieving relevant sections through high-performance vector similarity searches.

AgentDB vector search supports hybrid search by combining vector similarity with metadata filters. This allows you to execute precise queries that narrow down semantic search results based on specific document attributes.

To run a local vector database with HNSW indexing, you need Node.js 18+ and AgentDB v1.0.7 or higher. You also need an embedding model to generate the vectors for storage and retrieval.

You query a local vector database from the command line using npx agentdb, passing the database file, your embedding vector, and the desired number of results. This returns the most similar vectors using HNSW indexing and quantization.