AI NewsInfrastructureAnnouncement
Cloudflare adds a Web Search API to AI Gateway with Ceramic, Exa and Linkup, priced at the partner's own rate
Cloudflare's AI Gateway now routes web search calls to Ceramic.ai, Exa and Linkup from a single REST endpoint or Workers binding, priced at the partner's own list rate and billed out of AI Gateway credits.

Image: Cloudflare
Why it mattersAn agent that needs the live web no longer has to guess a URL and curl it, and the team no longer has to sign a separate contract per search provider to find out which one gives the best grounding.
An agent asked to find today's news usually does one of two awkward things: it guesses a URL and runs curl, which fails often enough that developers now treat 404s as a known failure mode, or the team wires in a search provider by hand and signs a separate contract to do it. Cloudflare announced a Web Search API through AI Gateway today that removes the second step and gives agents a cleaner way to do the first.
Cloudflare says the launch partners are Ceramic.ai, Exa and Linkup, and that the search provider is picked by name in each request. The API is reachable three ways: a direct REST endpoint under api.cloudflare.com, a Workers binding on env.AI.websearch, or as part of a model call through the existing AI Gateway chat completions path. The company says native server tools, where the gateway orchestrates the search call for the model, are coming next.
How it is priced
Cloudflare says search calls are billed from the same AI Gateway credit balance that already covers inference, and that the per-call rate is each partner's own list price with no markup added. Teams that already pay one of the partners directly can use Bring-Your-Own-Key, and Cloudflare says it will not insert itself into the billing in that case.
Cloudflare also says it will mark the partners that offer Zero Data Retention, so a team can tell at a glance which provider will keep a copy of the query. The company has not published a comparison of the three partners on recall or latency, so the choice of which one to use is still a job for the team to measure on its own traffic.
The crawler terms are part of the pitch
Cloudflare says every launch partner has agreed to meet its public rules for Verified Bots, which include identifying the crawler, respecting robots.txt and returning a link to the source for every result. That last clause is the one that matters most for a reader who wants to check what the agent based an answer on, and it is also the one many search APIs skip. Cloudflare has been building a case over the past year that answer engines should pay publishers for the content they cite, and this launch extends the same terms to the search layer that feeds those engines.
The company has not said what happens if a partner stops meeting those rules, or whether a team using Web Search API would be notified before the partner was removed.
Teams building retrieval into an agent now have a one-command path to try all three providers against the same prompt, from the same observability dashboard, without three contracts. Whether to keep the gateway in that path once a provider is chosen is the harder question, because every search call still rides on Cloudflare's billing.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.


