Cloudflare Launches Web Search API for AI Agents

AI agents have a search problem. They either hallucinate URLs, rely on stale training data, or depend on ad-supported search APIs that were never designed for machine consumption. Cloudflare's new Web Search API, launched October 2 as a beta, wants to fix that.[1]

The API runs through Cloudflare's AI Gateway and lets agents query the live web through one of three search providers: Ceramic.ai, Exa, and Linkup. All three support Zero Data Retention for requests routed through Cloudflare, and all have committed to Cloudflare's verified bot crawling standards. Search requests show up in your gateway logs and are billed to your AI Gateway credits at each provider's list price, with no additional markup. You can also bring your own provider API key.[1]

The launch hit 548 points on Hacker News, making it the most upvoted story of the past week. The discussion spanned 248 comments.[2]

How It Works

You call a REST endpoint with a query, a provider name, and a result limit. The request goes through AI Gateway, which handles logging and billing. From a Cloudflare Worker, you can also use the AI binding directly, no HTTP client needed. The response is structured JSON with search results ready for an agent to parse.[1]

This is not a search engine. Cloudflare is not crawling the web themselves. They are acting as the infrastructure layer between your agent and existing search providers, bundling access, logging, and billing into one API call. The value proposition is that you do not need to evaluate, integrate, and manage three separate search providers. Cloudflare does that, and you pay roughly what you would pay going direct.[1]

Why It Matters

Every agent framework worth its name has a web search tool bolted on. The quality of that tool determines whether the agent gives you a real answer or a confident-sounding fabrication. Until now, most agent builders either used Bing's API (now retired in favor of a Copilot-oriented product), Google's Programmable Search Engine (rate-limited and designed for embedding in websites, not agents), or one of the newer search APIs like Exa or Tavily.[3]

Cloudflare's move matters for two reasons. First, it standardizes the interface. If you build against the Cloudflare Web Search API, you can switch providers by changing one parameter. Second, it normalizes zero data retention. Search queries from agents are sensitive. They reveal what a user is working on, what they are confused about, what they are buying. Zero Data Retention means the provider cannot store or reuse those queries.[1]

The Catch

It is a beta. The provider list is three names long. Pricing is passthrough, which means it is only as cheap as the underlying providers, and none of them are free for serious use. And Cloudflare's AI Gateway, while functional, is still a relatively new product that has yet to prove itself at scale for production agent workloads.[1]

There is also the question of lock-in. The API is simple enough that migrating away would take an afternoon. But the logging, the gateway integration, the Workers binding, and the billing consolidation all make it stickier over time. Cloudflare is betting that convenience compounds.[1]

For agent builders who were already piecing together search infrastructure, this is a clean answer. For everyone else, it is a signal that the search layer of the AI stack is becoming a product category in its own right. The web is not going anywhere. The question is who builds the best pipe to it for machines that need to read it.[1]