Cloudflare launches Web Search API via AI Gateway with Ceramic.ai, Exa and Linkup

Cloudflare launches web search for agents through AI Gateway

Cloudflare has introduced a Web Search API delivered through AI Gateway, the control plane it offers for AI applications, with three search partners at launch: Ceramic.ai, Exa and Linkup. According to a Cloudflare blog post published on 2 October 2026, the feature targets AI agents that need current information rather than whatever a model happened to memorise during training.

The post describes a habit many agent developers will recognise: when a tool-using model wants a web page, it often invents a plausible address and issues a fetch, producing 404 errors whenever the guess lands wrong. Cloudflare's argument is that agents should behave more like people, beginning from a search query and working from genuine results, which both cuts wasted requests and grounds model output in live data.

Fresh context for frozen models

Language models are trained up to a cutoff point and then effectively frozen, the blog notes, which makes them shaky on recent events, evolving APIs and fast-moving news. Feeding search results into a model's context gives applications a fresh, structured layer of information pulled from the live web. Cloudflare offers a concrete example: an agent helping someone build with its developer tools would otherwise be unaware of products announced during the current Birthday Week, while a search step can surface the newest documentation and releases.

Crawler standards as a condition of partnership

The more unusual part of the announcement is the condition Cloudflare attached to its partners. The company says the participating providers have committed to its bot-crawling standards: their crawlers must meet the publicly documented requirements for Cloudflare's "Verified bots" programme, and every search response must include a link back to the location of the crawled content. Cloudflare presents the rules as a way for creators to keep control over their data, with crawlers that identify themselves, respect robots.txt and cite where results came from.

Integration, billing and what comes next

Search queries are consumed through AI Gateway, so they draw down the same credit balance used for model inference, appear in existing observability logs, and can be governed with access controls. Cloudflare says it will flag which partners support Zero Data Retention, that results are priced at the providers' list rates with no markup added, and that Bring-Your-Own-Key arrangements work for organisations that already hold provider accounts.

Two interfaces are available now. A REST endpoint accepts a query, a provider name and a result limit, authenticated with a Cloudflare API token, for use from backends, mobile apps or external services. On Workers, a binding exposes search as a single call that returns JSON results. Cloudflare is also building native server tools into AI Gateway, with web search among the first; until those ship, developers can wire search up themselves by declaring a web_search function tool, catching the model's tool call, running the search binding and passing the results back into the conversation.

Why it matters

The launch takes a fiddly, ad hoc piece of agent engineering, grounding answers in live web data, and turns it into a managed service with logging, billing and access controls in one place. It also clarifies how Cloudflare means to compete here: not by building its own search index, but by brokering partner indexes under its own crawler-transparency rules, an approach that could push source attribution and robots.txt compliance toward becoming baseline expectations for search APIs aimed at AI systems. For teams already routing inference through AI Gateway, the cost of adding grounded search to an agent is now close to zero.