- Published on
AI Gateway adds DeepSeek Search and Fetch tools
- Authors
- Name
- DeepHorus
AI Gateway has added DeepSeek Search and DeepSeek Fetch, two hosted tools for web research and page retrieval through one API key. Search returns relevant pages with titles, URLs, and summaries, while Fetch retrieves page content as Markdown for models to read. The tools work with any model supporting tool calling, and no separate DeepSeek account is required.
The release extends AI Gateway’s role as a unified layer for model access, usage tracking, cost accounting, and routing. Previously, teams seeking web search and page extraction often had to integrate separate providers and manage distinct credentials. AI Gateway now bundles those capabilities into its tool ecosystem, keeping the same tool definitions when developers switch model providers.
Developers can use the helpers in AI SDK 6.0.321 or later on v6, and 7.0.160 or later on v7. Installation uses pnpm add ai@latest, and applications set AI_GATEWAY_API_KEY. In code, deepseekSearch accepts a count parameter, while deepseekFetch retrieves a selected result URL.
The tools can be passed together to streamText, allowing a model to search, fetch, and summarize in one flow. Developers can also pair DeepSeek Fetch with alternative search tools such as Exa Search or Parallel Search. If DeepSeek Search fails, a model can try another search tool, then fetch a result URL.
Search and Fetch each cost $0.75 per 1,000 billable requests. Failed requests, timeouts, and empty results do not incur a tool charge, though model tokens are billed separately. Search fallback is model-directed, so provider order and retry limits are best effort rather than guaranteed.
The impact is broadest for teams building research agents, retrieval-augmented generation pipelines, and assistant workflows. They can add web discovery and page reading without a DeepSeek account, and they can change model providers without rewriting tool integrations. The single-key setup also simplifies billing and access control, while the fallback pattern may help production systems handle provider errors more gracefully. For developers, the appeal is a simpler path from prompt to sourced answer.
AI Gateway says configuration options and pricing details are documented for both tools. The addition fits a larger trend: gateway platforms are moving beyond model routing into bundled tools for search, retrieval, and agent workflows. That could make AI Gateways stickier as neutral control planes for multi-model applications.
