Cloudflare Adds Multi-Provider Web Search to AI Gateway
Developers can retrieve live results through REST or Workers while using gateway logging, billing and access controls across three initial search providers.
Edited by Tyronne Panaino
Cloudflare announced on October 2 that AI Gateway now includes a Web Search API, giving developers a single Cloudflare entry point for live search results from Ceramic.ai, Exa and Linkup. The company announcement says applications can call the service from a conventional backend or through a Cloudflare Workers binding, making the change relevant to teams building agents that need current web context.
The practical shift is not a new foundation model. It is a new retrieval layer inside the same control plane developers may already use for model inference. Cloudflare says search calls can appear in AI Gateway logs, consume gateway credits and follow gateway access controls. That could reduce the operational separation between model calls and the web searches that supply their context, although the fetched evidence does not establish that the combined workflow is more accurate or reliable.
One interface covers three initial search providers
The launch begins with Ceramic.ai, Exa and Linkup. A developer selects a provider in the request and receives structured search results that can be passed into a model's context. Cloudflare presents this as a way to retrieve current material rather than depend only on information available when a model was trained.
Two current access paths are documented in the announcement. Existing backends and mobile or external services can use a REST endpoint with an AI Gateway token. Applications running on Workers can use a dedicated binding. Cloudflare also shows how a team can orchestrate search as a tool in its own agent loop, but native server tools inside AI Gateway are described as future work rather than a currently shipped capability.
That distinction matters for implementation planning. Search is available now as an API or binding, while automatic server-tool integration remains on the roadmap. Teams adopting the release still need to decide when their agent should search, which provider to call, how to rank or filter returned results, and how to handle unavailable or contradictory sources.
Crawler commitments become part of provider selection
Cloudflare says each participating provider has committed to its verified-bot requirements and that search responses must include a link to the source location. It also says those crawlers should identify themselves and respect site owners' robots.txt preferences. These commitments give developers a stated provenance and crawling baseline when choosing among the initial providers.
The announcement does not include an independent audit of compliance with those rules. It also does not establish how quickly robots.txt changes propagate, how providers handle removed pages, or whether source links remain accurate after content changes. The launch therefore documents a provider requirement, not proof that every crawl and result has already been checked against it.
Billing, keys and data handling stay visible
Search requests draw from AI Gateway credits at the provider's listed API price, according to Cloudflare, without an added Cloudflare markup. Teams can also bring an existing provider key. Cloudflare says queries can be inspected in normal gateway logs and that provider access can be controlled through the gateway.
The company also says it will identify providers that support zero data retention. That wording leaves a provider-specific question for buyers: the announcement does not publish a complete retention matrix, regional availability table or service-level commitment. Teams handling sensitive queries should verify the selected provider's retention terms and test which request data appears in logs before treating the gateway as a complete governance boundary.
What developers still need to test
The source establishes the launch, initial providers, integration routes, billing model and stated crawler rules. It does not provide independent relevance benchmarks, latency measurements, failure rates or comparisons across Ceramic.ai, Exa and Linkup. It also does not show how search quality changes when the same query is routed between providers.
Useful next checkpoints are production documentation for limits and availability, provider-by-provider retention disclosures, evidence that crawler commitments are monitored, and independent tests of freshness, source attribution and result quality. Until then, the release expands Cloudflare's developer surface but does not by itself prove better factual accuracy for the agents that use it.
Status
Confirmed. Cloudflare's official announcement establishes the October 2 launch and its current product scope. Internal confidence is medium because all capability and control details come from Cloudflare, with no independent operational testing in the fetched evidence.
Sources
Update note: Last reviewed 2026-10-08. We will revise this post if Cloudflare publishes provider limits, retention details, independent performance evidence or native server-tool availability.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.