Search and retrieval

Vectara provides a powerful, end-to-end search that retrieves relevant answers from your data. You can start with a simple query and then refine it with advanced features like hybrid search, reranking, and metadata filtering to achieve pinpoint accuracy.

Start here

• Quickstart: Run your first query against a corpus.

  Start here if you have data indexed and want to see results.

Refine what gets retrieved

Shape which results come back, and in what order.

• Tune retrieval: Adjust how results are selected and scored.
• Hybrid search: Combine semantic and keyword matching in one query.
• Reranking: Reorder results after retrieval. Covers the available rerankers, chaining them, and cutoffs.
• Filters: Narrow results using metadata filter expressions.
• Fuzzy matching: Match metadata values that are close but not exact.
• Intelligent query rewriting: Reshape an incoming query before it runs.
• Custom dimensions: Influence scoring with your own numeric dimensions.

Generate answers from results

Turn retrieved results into a response.

• Vectara prompt engine: Control how the prompt sent to the model is constructed.
• Custom prompts with metadata: Use document metadata inside your prompts.
• Citations: Configure how sources are cited in generated answers.
• Bring your own LLM: Use your own model for generation.

Put it in front of users

• Build a search UI: Build a search interface on top of these APIs.