Traditional databases index exact values and ranges. A vector store instead builds an index over high-dimensional vectors, typically using approximate nearest neighbour algorithms such as HNSW or IVF, so it can return the top matches from millions of entries in milliseconds without comparing against every record. Most also support metadata filters so results can be restricted by source, language or date.
In an AI assistant, the vector store is the retrieval layer. Source content is chunked, embedded and inserted once; at question time the query is embedded and the nearest chunks are fetched and passed to the language model as grounding context. Keeping the store current is what keeps answers current.
Clerkzo maintains a vector store per site, populated by crawling the customer's own pages. When the site changes and is re-crawled, the stored vectors are refreshed, so the widget answers from the live version of the business rather than a stale snapshot.
Related terms
Browse all 142 terms- AI hallucinationAn AI hallucination is an output from a language model that is fluent and confident but factually wrong or entirely invented. It happens because the model generates plausible language rather than retrieving verified facts.
- Semantic searchSemantic search finds content based on meaning rather than literal keyword overlap. It converts both the query and the indexed content into embeddings and returns the items that are closest in meaning, even when no words match.
- Web crawlerA web crawler is a program that automatically browses websites, fetching pages and following their links to discover more pages. Search engines, archives and AI systems use crawlers to collect the content they later index or answer from.
- Shadow DOMThe Shadow DOM is a browser standard that attaches a separate, encapsulated DOM tree to an element. Styles and markup inside that tree are isolated from the rest of the page, so neither side can accidentally restyle or select the other.