Open Brain
VercelEngineering PostOfficial Source

AI Gateway adds DeepSeek Search and Fetch tools

AI Gateway now exposes DeepSeek Search and Fetch to tool-calling models with one gateway key. Each costs $0.75 per 1,000 billable requests, with model tokens billed separately.

Vercel · Nov 6, 2026
Open Source Open MarkdownOpen JSON
Source Summary

DeepSeek Search returns page titles, URLs, and summaries; DeepSeek Fetch retrieves a result page as Markdown. Both are available through AI Gateway to any model that supports tool calling, using one gateway key. They require **AI SDK 6.0.321+ on v6** or **7.0.160+ on v7**.

Practical Implication

Builders can add the helpers to an agent’s tools and keep them when switching model providers. Search can be paired with Fetch, which can also retrieve URLs returned by alternative search tools.

Agent-Ready Context
DeepSeek Search returns page titles, URLs, and summaries; DeepSeek Fetch retrieves a result page as Markdown. Both are available through AI Gateway to any model that supports tool calling, using one gateway key. They require **AI SDK 6.0.321+ on v6** or **7.0.160+ on v7**.

Builders can add the helpers to an agent’s tools and keep them when switching model providers. Search can be paired with Fetch, which can also retrieve URLs returned by alternative search tools.

Each tool costs **$0.75 per 1,000 billable requests**; failed, timed-out, or empty requests are not charged, and model tokens cost extra. Search fallback order and retry behavior are model-directed and best effort.
Connected Context · Feed7 Judgment

This adds a paired search-and-fetch path to AI Gateway: agents can find pages and retrieve them as Markdown through tools usable with tool-calling models. It reinforces retrieval as swappable gateway plumbing while narrowing portability to supported AI SDK versions and adding per-request charges; the brief provides no retrieval-quality evidence.

Context Map
toolsresearchcoding#tool-use#agent-sdks#gateways
Uncertainty
Each tool costs **$0.75 per 1,000 billable requests**; failed, timed-out, or empty requests are not charged, and model tokens cost extra. Search fallback order and retry behavior are model-directed and best effort.