Cloudflare has introduced a beta Web Search API through AI Gateway, InfoQ China reported on November 11, 2026. The API can be reached through a REST endpoint and Cloudflare Workers. It provides unified logging, access control, and billing, and it supports bring-your-own provider keys.
The API lets AI agents retrieve live web search results from Ceramic.ai, Exa, and Linkup. Its unified interface folds search billing into AI Gateway, so search requests and model inference can be logged together. Providers must follow verified-bot requirements and return links to sources.
Cloudflare says the API allows AI applications to retrieve current web content, including recent news, API changes, and technical documentation, and to supply those results to a language model as context during inference.
Group product manager Michelle Chen, senior director Sam Else, and senior systems engineer Gabriel Massadas described the problem the company is trying to address. When an agent needs to fetch a page in real time, it often guesses the URL and then calls a tool that retrieves the page with curl. That is why a fetch sometimes returns a 404 Not Found: the agent guessed the wrong URL. Always guessing URLs at random, they said, is not a particularly efficient approach.
Applications and external services can use the REST endpoint by supplying an AI Gateway authentication token and naming the search provider in the request payload.
Chen, Else, and Massadas added that crawlers used by web search providers must meet the public “verified bots” requirements Cloudflare defines in its developer documentation, and that web search responses must include links to the crawled content. They said these rules support a fairer internet and let content creators decide how their own data is handled.
In a popular Hacker News thread, commenters discussed search quality, limits on data retention, and whether a unified API has a substantive advantage over existing alternatives.
Simon Willison wrote that his first question about a search API is whether it lets you store and republish results obtained from it. If he were running an agent system and were not allowed to store responses—or to offer a button to share a conversation transcript—that would be a significant limitation. He said the answer is often buried in the terms of service.
Maximilian Goisser, chief technology officer and co-founder of Erato Labs, said that for enterprises already using Cloudflare, the company is gradually becoming a trusted primary cloud provider, a role usually held by AWS, GCP, or Azure. Most of an enterprise’s cloud spending would then be settled through that platform, so a new procurement process would not be required.
Cloudflare said it will select providers that support “zero data retention” and will offer search at each provider’s list price, with no extra fee. The company plans to add native server tools to AI Gateway, starting with web search. Until then, developers can use Cloudflare Workers to integrate web search into their workflows.
The InfoQ China report identifies this English InfoQ article as its source: https://www.infoq.com/news/2026/10/cloudflare-web-search-api/
Original article: https://www.infoq.cn/article/SgXuVvzbHr8lKrihTAZW