Cloudflare launched a Web Search API inside its AI Gateway Friday, letting developers feed live web results into model calls through three search partners: Ceramic.ai, Exa and Linkup. The company announced the product in a blog post on the last day of its annual Birthday Week, pitching it as a way to give AI agents information that postdates their training data.
What Happened
The API gives agents and applications a single interface to the three providers. Developers can call it through AI Gateway, Cloudflare's control layer for routing and monitoring model traffic; through a REST API with an authentication token; or through a Workers binding that Cloudflare says takes one line of code. A fourth option, native "server tools" that would let a model invoke search without the developer orchestrating the tool call, is listed as coming soon.
Billing runs through existing AI Gateway credits. Cloudflare says queries are charged at each partner's list API price with no markup, and that search requests appear in the same observability logs as model requests. Customers who already have contracts with a provider can bring their own API keys, as they can for model providers on AI Gateway. Cloudflare will mark which partners support zero data retention, meaning queries are not stored by the provider.
Partners also have to meet conditions. Cloudflare requires them to comply with its verified-bots specification, which means identifying their crawlers, to respect robots.txt directives and to include links to sources in the results they return.
The company framed the launch around a basic limitation of language models. In the post, it argued that models are only as good as the context supplied to them, and a model trained on a fixed snapshot cannot answer questions about recent events, prices or documentation changes without retrieval.
Why It Matters
Web search has become a standard component of agent stacks, and developers typically wire it up separately from their model provider, with separate keys, bills and logs. Folding search into AI Gateway lets Cloudflare capture more of each agent request and gives customers one place to set spending limits and inspect what an agent fetched.
The partner rules matter as much as the plumbing. Cloudflare sits in front of a large share of websites and has spent the past year positioning itself between AI companies and publishers, including blocking AI crawlers by default for new customers, its pay-per-crawl program and, as The Primary Wire reported Wednesday, a test of paying publishers per AI citation. Requiring search partners to declare their crawlers and obey robots.txt means traffic sold through Cloudflare's own product follows the same norms it asks other AI firms to adopt. Source links in results also give publishers a path to referral traffic.
For Exa, Linkup and Ceramic, the integration provides distribution to Cloudflare's developer base, while putting their prices side by side in one catalog where customers can switch providers.
What to Watch
Key questions are whether Cloudflare adds more search providers, and whether large search operators join under the same crawler conditions. The timing of the server tools feature will determine how much orchestration code developers can drop. Publishers will also be watching whether the citation requirement produces measurable referral traffic, and whether Cloudflare connects search usage to its publisher payment experiments.
TECHNOLOGY · AI · CYBERSECURITY · THE SYSTEMS SHAPING BUSINESS