Vercel adds DeepSeek Search and Fetch to AI Gateway

November 6, 2026Source: Vercel

Vercel has added DeepSeek Search and DeepSeek Fetch to AI Gateway, allowing developers to give models web search and page retrieval tools through one AI Gateway API key. The tools are available with models that support tool calling, without requiring a separate DeepSeek account.

Key points

DeepSeek Search and DeepSeek Fetch are now available through Vercel AI Gateway.
Search returns page titles, URLs, and summaries, while Fetch returns page content as Markdown.
The tools work with any model that supports tool calling and one AI Gateway API key.
Each tool costs $0.75 per 1,000 billable requests.
Failed requests, timeouts, and empty results have no tool charge.

Search and fetch through one gateway

DeepSeek Search finds relevant pages and returns their titles, URLs, and summaries. DeepSeek Fetch retrieves the content of a page as Markdown, giving a model material it can read after a search result has been selected.

Vercel says developers can use the two tools with any model that supports tool calling and keep the same tools when switching between model providers. AI Gateway also provides one API for calling models, tracking usage and cost, and configuring routing, retries, and failover.

The integration is available in AI SDK 6.0.321 and later on v6, and in AI SDK 7.0.160 and later on v7. Developers can update with pnpm add ai@latest and set the AI_GATEWAY_API_KEY environment variable.

Fallback options for search failures

Vercel's example combines deepseek_search and deepseek_fetch in the tools passed to streamText. The example uses the model deepseek/deepseek-v4.1-flash-1017 to find the latest Vercel AI Gateway announcement, retrieve it, and summarize it.

Developers can also configure alternative search tools if DeepSeek Search fails. Vercel's second example uses exa_search and parallel_search alongside DeepSeek Search, then uses DeepSeek Fetch on a result URL after a search succeeds.

The fallback is model-directed, so provider order and retry limits are best effort, according to Vercel. DeepSeek Fetch can retrieve a result URL returned by any of these search tools. This gives applications a way to retain a page-retrieval step while allowing the model to try other configured search providers.

Usage charges and developer impact

DeepSeek Search and DeepSeek Fetch each cost $0.75 per 1,000 billable requests. Failed requests, timeouts, and empty results have no tool charge, while model tokens are billed separately. Vercel directs developers to the DeepSeek Search and DeepSeek Fetch documentation for configuration options and pricing details.

For developers building research, browsing, or summarization workflows, the update reduces the need to manage a separate DeepSeek account for these tools. The ability to switch model providers without changing the tools can also simplify routing and failover configurations, while the stated per-request charges provide a separate cost category from model token usage.