Classic search systems (full-text search, SQL) find documents that contain exactly the searched terms. A vector database works differently: each document is converted into a high-dimensional numerical vector by an AI model (embedding). Similar content is geometrically close in vector space.

When a search query arrives, the query is also converted into a vector, and the system finds documents whose vectors are most similar to the query — regardless of whether exactly the same words are used. This enables semantic search: a question about ​‘vacation policy for remote work’ will also find documents using the terms ​‘home office’ and ​‘time off in lieu’.

In a RAG architecture, the vector database is the central component: it contains all indexed enterprise documents as embeddings and delivers the most relevant text passages to the language model when a question arrives. Silent AI includes an integrated vector database — no external system is required.

Frequently asked questions

No. Silent AI contains a fully integrated vector database. It is automatically populated when the connectors (SharePoint, Confluence, file servers, etc.) ingest documents. No external database product, no separate installation.
Connectors synchronize the dataset at configurable intervals. New or changed documents are automatically re-indexed. The degree of currency depends on the synchronization frequency and is configurable.