Cloudflare adds Web Search API to AI Gateway with three launch partners

Why agents need search instead of guessing

Cloudflare describes a common weakness in how agents fetch live pages: they usually guess the URL and make a tool call to curl it. When the guess is wrong, the result is a 404 Not Found. Cloudflare argues that a better approach is for agents to search the way people do, starting with a search engine query.

The company also points to the knowledge cutoff problem. Models are typically trained and then frozen at a point in time, which makes it hard to discuss recent events, changing APIs or fast-moving news. According to Cloudflare, integrating the Web Search API into an inference pipeline gives agents a dynamic context layer, with fresh, structured snippets from the web injected straight into context.

As an example, Cloudflare says an agent building with its developer tools could otherwise miss the products and features released during Birthday Week. With web search, it could retrieve the latest documentation and releases.

Crawler standards for search partners

Cloudflare says its partners have committed to meeting its bot crawling standards. The crawler used by each web search provider must comply with Cloudflare's publicly stated requirements for "Verified bots", as defined in its developer documentation. Web search responses must also include a link to the location of the crawled content.

The company says users who choose web search on Cloudflare consume knowledge from operators that identify their crawlers, respect robots.txt and provide the source of search results. Cloudflare frames this as a way to give site owners transparency and control over how their content is used.

How it works in AI Gateway

Cloudflare calls AI Gateway the flagship integration point for the new product and describes it as a control plane for applications, with observability, unified billing, security and access controls built in. For web search, that means:

• Web search can be paid for with AI Gateway credits.
• Requests show up in normal AI Gateway observability logs.
• Teams can control who gets access to which web search providers.

Cloudflare says it will identify partners that support Zero Data Retention (ZDR), so users know their data is not retained. Web search is offered at the partners' list API pricing, without any additional markup. Bring-Your-Own-Key (BYOK) is supported with search providers, as it is with model inference providers.

REST endpoint, Workers binding and server tools

Developers can call the service in two ways. Backends, mobile apps or external services can use a standard REST endpoint, passing an AI Gateway authentication token and naming a provider in the payload. The example in the post uses the providers ceramic and exa, with a query and a result limit of 5. For those building on Cloudflare Workers, a standalone worker binding exposes a websearch call from the AI environment object.

Cloudflare is also building native Server Tools into AI Gateway. The idea is that developers will no longer need to define tools themselves, because they will come built into the control plane. Web search will be one of the first. Until then, the post includes a Worker snippet showing how to orchestrate web search as a tool yourself, using the model @cf/google/gemma-4-26b-a4b-it.

What it means for developers

For teams already running agents through AI Gateway, the change adds live web grounding without a separate billing or logging setup, since search usage appears alongside other gateway activity. The partner requirements on crawler identification and source links also give developers a way to choose search providers that meet Cloudflare's published rules. Cloudflare says users can get started through its developer docs or try the feature in the AI Playground with an AI Gateway account and key.