Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Google brings Tool Retrieval to the Gemini API for agents with hundreds of tools

Google has added Tool Retrieval to the Gemini API, now available in preview for developers building agents. The company says the feature lets agents reach hundreds of tools while spending fewer tokens on every request.

Published
Gemini API 上线工具检索:让智能体接数百个工具、少花 token
Image source: gemini.google

Google has added a capability called Tool Retrieval to the Gemini API, the company said in a post on its developer blog. The feature is available in preview, which means developers can call it now while the interface and behaviour may still change.

The pitch is built on two constraints: scale and cost. According to Google, Tool Retrieval gives agents access to hundreds of tools while using fewer tokens. That combination matters because tool definitions are not free: every tool an agent can call has to be described to the model, and those descriptions consume context on every request whether or not the task needs them.

Placing the announcement in a developer tools section signals how Google frames it — as part of the plumbing for building agents rather than a product update end users will notice.

For builders, the feature addresses the selection problem. Once an application connects to calendars, mail, databases and internal endpoints, the model has to pick the right action at the right moment out of a large library. Tool retrieval shifts that step into a retrieval layer, so the model works from a relevant subset instead of the entire catalogue.

That is also why token count sits in the headline. Context is one of the most direct costs in running an agent, and tool descriptions are a passive contributor to it, repeated on every call even when the current task touches none of them.

Google has not disclosed the maximum number of tools supported, how the retrieval step is implemented, or how the capability is priced. Those details will matter to teams deciding whether to build a production system on top of a preview feature.

What to watch next is whether Tool Retrieval stays a standalone API capability or gets bundled with the rest of Gemini's agent tooling, and how quickly third-party tools organise themselves around it. Tool count is becoming a dividing line in agent products: demonstrating a handful of tools is easy, running hundreds of them in production tests retrieval quality, permissions and failure handling at once.

Why it matters

Tool retrieval moves the scaling limit for agents from the context window to retrieval quality, which changes the cost structure of multi-tool applications. Its general availability and pricing will decide whether hundred-tool agents become ordinary products.

GoogleGeminiAgentAPI
Back to realtime news

Nearby Updates

All