Cloudflare on October 2, 2026, launched an open beta of its Web Search API through AI Gateway, offering developers structured web index access designed for language-model retrieval pipelines. It returns titles, URLs, and text descriptions from external search providers for applications to pass into language models.
The service launches with three supported search providers: Ceramic.ai as the default engine, alongside Exa and Linkup. The API returns search results; applications handle the model calls and use those results to inform answers.
Integration, Access, and Permissions
Developers can access the API either through standard REST endpoints or natively within Cloudflare Workers using the websearch() method on the Workers AI binding. Setup requires an active Cloudflare account, an AI Gateway, and either loaded AI Gateway credits or a provider API key stored on the gateway. Every account has a gateway named default; developers can also create another.
For REST implementations, Cloudflare requires API tokens to hold both Workers AI > Read and AI Gateway > Read permissions. Applications can fund queries using Cloudflare prepaid credits or connect their own provider keys to route charges directly to an existing account with Ceramic, Exa, or Linkup.
Cloudflare says native server tools are coming soon. For now, developers define a search tool, execute its calls and pass the returned results back to the model.
Pricing and Provider Policies
Cloudflare passes third-party query costs through at provider list prices without adding a platform markup. When developers configure their own provider API keys, billing occurs directly through the respective search vendor.
Cloudflare’s October 5 documentation snapshot lists the following prices and retention policies:
| Provider | Cost per 1,000 queries | Stated zero-data retention | Default selection |
|---|---|---|---|
| Ceramic.ai | $0.25 | Yes | Yes |
| Exa | $7.00 | No | No |
| Linkup | $5.00 | Yes | No |
The retention entries describe provider policies listed by Cloudflare. Its announcement says the providers commit to verified-bot requirements, including respecting robots.txt and linking to original sources; these commitments do not establish independent verification of every crawl.