AgentDB Vector Search

Semantic vector search improves RAG retrieval by storing documents as embeddings and matching meaning rather than exact keywords. This finds relevant context even with different terminology, directly improving downstream LLM answer quality.

To initialize a vector database, use the local CLI to set up your store and insert embedded documents via API. Query the database by specifying retrieval parameters like k, distance metric, and threshold to return the top ranked similar matches.

Yes, you can filter vector search results using hybrid metadata filtering. This combines vector similarity with metadata filters like domain, category, and date ranges to return highly targeted results from your embedded document collections.

MMR (Maximal Marginal Relevance) is a context selection technique that ensures diversity in retrieved documents. Use MMR during RAG-ready querying to synthesize varied, useful context and avoid redundant information for downstream LLM answers.

Yes, AgentDB vector search achieves fast scalable retrieval at scale. It uses high-performance HNSW indexing, quantization, and caching to deliver sub-millisecond pattern retrieval across large collections of embedded documents.

You can use cosine similarity, euclidean distance, and dot product as distance metrics for similarity matching. Selecting the correct metric alongside proper vector dimensionality ensures your embedding pipeline returns accurately ranked results.