What is a vector database?
A vector database is a data store designed to hold embeddings and quickly find the ones most similar to a query, which makes it the usual search layer behind AI assistants that answer questions from a company's own documents.
A normal database is good at exact matches: find the customer with this phone number. A vector database is good at nearest matches: find the ten passages that mean something closest to this question. Products such as Pinecone, Weaviate and Qdrant specialize in this, and mainstream databases like PostgreSQL can do it with the pgvector extension.
Most small and mid-sized businesses never need to choose one directly. Off-the-shelf AI assistants, Microsoft 365 Copilot and many help desk tools manage this layer behind the scenes. You start caring when you commission a custom knowledge assistant, because then someone must decide where the vectors live, how they stay in sync with source files, and who can query them.
If a build includes a vector database, ask practical questions. Does it respect document permissions, so staff only retrieve what they are allowed to see? How is it updated when a file changes or is deleted? Where is it hosted, and is the data encrypted? Deleting a client's files should also remove their embeddings, which is easy to forget.
Example
For example, a regional home health agency building an internal assistant for care coordinators could store embeddings of its policies and care protocols in a vector database. Because some documents reference patient scenarios, the agency hosts it within its HIPAA-covered environment, restricts access by role, and logs every query. A clinical lead reviews answers during the pilot.
Where it shows up in practice
See how InTheory provides this: fractional CTO · fractional CIO · pricing.