Exa

Search API that finds pages by meaning for AI applications

AI Search & KnowledgeFree planOverseasβ˜…β˜…β˜…β˜…β˜† 4.0

What is Exa?

Exa is a search engine and API intended to be called by software rather than browsed by a person. Queries are matched by embedding similarity instead of keywords, so a request such as a blog post by someone learning to sail returns pages that read that way even when the words never appear. Endpoints also fetch cleaned page text ready to drop into a prompt, and find similar links from a single known URL. It is the search step developers bolt onto agents and retrieval pipelines, and pricing is usage based, so cost scales with how often an application searches.

Last updated: 2026-09-20. This site only provides an index; for exact features, pricing, and licensing, see the official website.

Key features

  • Neural search endpoint that ranks results by semantic similarity
  • Contents endpoint returning cleaned page text for prompt use
  • Find-similar endpoint that expands from one known good URL
  • Keyword and filtered modes for when exact matches matter
  • Python and JavaScript SDKs alongside plain REST calls
  • Domain, date and category filters on search requests

Pros & cons

Strengths

  • Semantic matching finds pages a keyword search never returns
  • Content extraction removes a scrape and clean step in RAG
  • API-first design fits agents and pipelines without extra glue

Watch out for

  • Usage-based pricing needs watching on high-volume applications
  • Result quality varies by domain, with news and blogs strongest
  • Index coverage is not the same as a general web search engine

Best for & use cases

rag pipelines, ai agents, semantic search and content retrieval

If you're comparing similar products, check the alternatives below, or browse all tools in the AI Search & Knowledge category.

FAQ

How is this different from a normal search API?

It ranks by meaning rather than keyword overlap and can return page text with the results. That combination lets an agent search once instead of searching, scraping and cleaning in separate steps.

Do I need to crawl the pages myself?

Usually not. The contents endpoint returns extracted text for the results, which removes most of the fetching and parsing work from a retrieval pipeline, though you still own chunking and ranking.

Is there a monthly free allowance?

Yes, a limited free credit allowance exists for evaluation. Past that, pricing follows usage, so prototype with realistic queries before estimating what production traffic will cost.