Cloudflare has added a native Web Search API to its AI Gateway, built in partnership with search providers Ceramic.ai, Exa, and Linkup. The feature lets developers pull live web results into a model's context window during inference, addressing the gap where models are frozen at a training cutoff and can't speak to recent events or changing documentation.
Three access paths are available: directly through AI Gateway, via a standard REST endpoint (passing an AI Gateway auth token and a provider name in the request payload), or through a Workers binding that Cloudflare says takes one line of code to call. Requests appear in the same AI Gateway observability logs as model inference calls, and usage draws down from the existing AI Gateway credit balance. Cloudflare says it is offering the search at list API pricing from partners with no added markup, and supports Bring-Your-Own-Key setups for organizations with existing provider accounts. The company also says it will flag which partners support Zero Data Retention so users know their queries aren't being stored.
Crawling behind the search results is tied to Cloudflare's bot policy: partner crawlers must meet Cloudflare's published "Verified bots" requirements, and every web search response must link back to the source of the crawled content.
Cloudflare also previewed upcoming "Server Tools" built directly into AI Gateway, which would remove the need for developers to define tool-calling logic themselves; web search is slated to be among the first tools included. Until then, Cloudflare published a Workers code sample showing how to wire web search into a tool-calling loop manually.