Vercel AI Gateway adds DeepSeek Search and Fetch under a single API key

Vercel has added DeepSeek Search and DeepSeek Fetch to its AI Gateway, giving developers web search and page retrieval through a single API key. According to the Vercel blog, both tools work with any model that supports tool calling, and no separate DeepSeek account is needed.

What each tool does

DeepSeek Search queries the web and returns matching pages, each with a title, URL and summary. DeepSeek Fetch takes a URL and returns the page's content as Markdown that a language model can ingest. Together the pair covers the standard pattern of finding a source online and then pulling its contents into a model's context.

Because the tools are exposed through the Gateway rather than tied to DeepSeek's own models, Vercel says they keep working when you switch model providers. A team can move its application from one provider's model to another without re-plumbing the search and retrieval layer.

Setup in the AI SDK

The helpers ship in AI SDK 6.0.321 and later on the v6 line, and 7.0.160 and later on v7. The update path Vercel describes is one command, pnpm add ai@latest, plus setting the AI_GATEWAY_API_KEY environment variable. To combine search with retrieval, developers pass both helpers into a model's tools configuration.

Vercel also allows mixing search providers. If DeepSeek Search fails, an alternative search tool can be configured as a fallback, and DeepSeek Fetch can pull a result URL produced by any of those search tools. There is a caveat: fallback selection is model-directed, so provider ordering and retry limits operate on a best-effort basis rather than as firm guarantees.

Pricing and failure handling

According to the Vercel blog, Search and Fetch each cost $0.75 per 1,000 billable requests. Requests that fail, time out, or come back empty are not billed at the tool level, which lowers the cost of running retry-heavy patterns in production. Model tokens are billed separately, so the total cost of a search-augmented call depends on the model doing the calling.

Where the Gateway fits

AI Gateway serves as Vercel's single interface for invoking models, metering usage and spend, and configuring routing, retries and failover. Customers can either use an AI Gateway API key or bring their own provider keys. The new tools extend that surface from model access into the surrounding utilities that grounded AI applications need.

Why it matters

Web search and retrieval are becoming baseline capabilities for AI products, and wiring them up has usually meant juggling separate vendor accounts, keys and billing. By folding search and fetch into the Gateway's single-key model, Vercel removes that friction and makes the retrieval layer portable across providers, a meaningful property as teams increasingly treat models as swappable components behind routing and failover setups.

The billing design also matters for reliability work. Since failed, timed-out and empty requests are not charged at the tool level, fallback chains and retry logic, exactly the kind of behavior AI Gateway is built to configure, will not silently inflate costs. Model-directed fallback is worth watching in practice given its best-effort nature, but the overall shape is a natural extension of what a gateway is for: one API, one bill, and provider choice behind the scenes.