A vector database indexes embeddings, which are numerical representations capturing the semantic content of text. Retrieval works by converting a query into the same representation and finding the passages closest to it in that space.
It is the storage layer that makes semantic search and retrieval-augmented generation practical at scale.
Alternative Names:
Vector Store, Embedding Database
Why it Matters?
Retrieval quality is usually the limiting factor on accuracy in matter-specific legal AI, and the vector layer is where that quality is determined. A system that fails to retrieve the relevant page cannot answer correctly regardless of how capable the model is. This is also where tenant isolation matters, since embeddings derived from one client's documents must not be searchable from another matter.
Frequently Confused with
Related terms
Frequently asked questions
Why not use ordinary keyword search?
Do embeddings contain the original text?
