Cloudflare puts web search inside AI Gateway

Cloudflare launched a Web Search API inside AI Gateway with launch partners Ceramic.ai, Exa and Linkup. Search is billed with AI Gateway credits, shows up in normal Gateway observability logs, and carries per-provider access controls. It is reachable three ways: through AI Gateway itself, through a direct REST endpoint at POST /client/v4/accounts/{id}/ai/websearch/, and through a Workers binding at env.AI.websearch({...}).

The differentiated part is not the product, it is the governance. Cloudflare is making crawler compliance a condition of partnership: partner crawlers must meet Cloudflare’s Verified bots requirements, must respect robots.txt, and web search responses must include a link to the crawled location. Cloudflare will also flag which partners support Zero Data Retention, which is a reasonable signal for buyers who need to know whether a query can be retained.

Pricing is at partner list API pricing with no additional markup, and BYOK is supported for both search and inference providers. Native Server Tools in AI Gateway are announced as coming, with web search slated to be one of the first tools shipped natively, so today you have to define the tool yourself.

The motivation Cloudflare gives is agent staleness. The framing is that agents otherwise miss fast-moving documentation and releases because of knowledge cutoff dates, which is the same problem the MCP wave is trying to solve from the other direction.

A note on provenance: the date is not printed in the visible page body. It comes from Cloudflare’s own developer changelog entry for the launch and from structured data embedded in the post, both of which read October 2, 2026.

(Original source: Cloudflare)