Cloudflare added a Web Search API to AI Gateway on 2 October through partnerships with Ceramic.ai, Exa and Linkup, giving developers a way to supply models with current information during an inference request. In its announcement, Cloudflare said searches can run through AI Gateway, a separate REST endpoint or bindings for its Workers platform. The company says the service charges partners’ list API prices without an additional markup.
Key points
- Ceramic.ai, Exa and Linkup provide searches through Cloudflare’s new API.
- Queries can use AI Gateway credits, and requests appear in Gateway logs.
- Partners have committed to crawler standards, and responses must link to crawled content.
Web Search API supplies current context
Cloudflare’s case for adding search starts with a constraint of model training: a model generally draws on information available before its knowledge cut-off, while documentation, releases and other material on the web can change afterwards. The company says its API retrieves structured snippets from current web results for an application to put into the context supplied to a model. That gives the model material to use when answering about something newer than its training data.
The distinction is much like answering from a book printed some time ago after consulting a current index. Search finds material relevant to the question; the application then passes that material to the model for its answer. Cloudflare’s example is an agent using its developer tools that needs information about newly released features. Its published Worker code shows a model requesting a search, the Worker fetching results and a further model call receiving those results as a tool response. The search and the answer are separate operations in that example.
Looking up instructions for a newly released feature could therefore involve an answer informed by a current search result rather than only by information available when the model was trained. The required link to the crawled content would provide a place to check the material behind that result.
AI Gateway handles searches and billing
For developers already routing model requests through AI Gateway, Cloudflare says web searches enter the same system for logs, billing and access controls. Search requests appear in its observability logs, queries draw down the Gateway credit balance, and access to search providers can be controlled there. The company offers the searches at each partner’s list API price, with no additional Cloudflare markup. It also supports Bring-Your-Own-Key, allowing an organisation to use an existing key for a search provider.
Cloudflare also provides a direct REST endpoint for software making HTTP requests. Its example sends an authentication token, a search query and a chosen provider in the request, along with a limit on returned results and a Gateway identifier. For applications built on Workers, the company shows a binding that accepts a query, provider and Gateway identifier and returns a response the Worker can read as search results.
Those routes matter to the engineering of the feature: an application can call search directly, or put it alongside model inference within a Gateway-managed workflow. In Cloudflare’s illustrated Worker sequence, the application still defines the search tool offered to the model and handles the model’s request to use it. Cloudflare says it is building native Server Tools into AI Gateway and intends web search to be among the first tools incorporated there.
Ceramic.ai, Exa and Linkup commit to crawler rules
The three search partners have committed to meeting Cloudflare’s standards for crawlers, according to the company. Cloudflare says a provider’s crawler must comply with its publicly stated requirements for Verified bots, including identifying itself and respecting robots.txt, the file websites use to express rules for automated visitors. Search responses must also contain a link to the location of the crawled content.
That link concerns the origin of a result, rather than the text of an answer produced from it. For a developer deciding how to present search-backed output, it provides a route back to the material the search provider found. Cloudflare describes the crawler commitments as a way for website owners to retain visibility and control over automated access to their content.
Cloudflare also says it plans to identify which search partners support Zero Data Retention, under which data is not retained. The Web Search API is available through AI Gateway and the standalone REST API, and Cloudflare says it can be tried in the AI Playground with an AI Gateway account and key.
.jpg){kind=link}