Exa
Exa is the search API to reach for when the consumer is a model rather than a person: it hands back page text and excerpts an LLM can use directly, and the Instant tier is fast enough to sit inside a chat turn. The pay-as-you-go pricing is cheap at the entry and transparent, but the menu is long, and the Deep and Agent modes can cost a hundred times a plain search, so a prototype that upgrades its search mode casually gets a surprising bill. It is not a product for anyone who wants a search box; there is no end-user app, only a dashboard and the API.
Pros
- Returns full page text or relevant excerpts, so results are usable by a model without a second crawl
- $10 of free credit every month with no card, enough to prototype on
- Instant tier with roughly 180 ms median latency for in-conversation search
- Deep search, Answer, Monitors, Websets and an Agent endpoint cover research jobs beyond one query
- SOC 2 Type II, HIPAA with a BAA and zero data retention options for regulated buyers
Cons
- Pricing has many line items: results beyond ten, summaries, content types and agent compute are each metered separately
- Deep and Agent modes cost $12 to $1,000 per thousand requests, so mode choice drives the bill more than volume
- Its own index is smaller than a general web engine's, so recall on obscure or very recent pages varies
- Headquarters and founding details are not stated on the site