Google previews Tool Retrieval in the Gemini Interactions API
Original title: Introducing Tool Retrieval in the Gemini API
Agents can defer tool schemas and load them on demand, cutting input tokens on Gemini 3.6 and later, including Gemini 4.
Google launched Tool Retrieval in preview in the Interactions API. Developers add a tool_search tool and set defer_loading on functions so the model sees names and descriptions but not full parameter schemas until it requests a tool by name. Search can run on Google's servers in the same request or on the client against a custom registry, and it also covers remote MCP servers. There is no separate fee; tokens are billed at the model's standard rates.
- Prefix caching stays stable because deferred schemas are added later in the conversation.
- Client-side search adds one round trip and can expose new tools mid-conversation.
- Turning on tool_search adds a small fixed block of input tokens.
- Loaded schemas are billed as input for the rest of the conversation.
- The sample request uses the gemini-4-hydrogen model.
- Vendors
- Products
- Gemini APIInteractions API
- Organizations
- Google DeepMind
- AI models
- Gemini 3.6Gemini 4gemini-4-hydrogen
From the original article
4 min read at blog.google
Give your agents access to hundreds of tools while using fewer tokens with Tool Retrieval, now in preview in the Gemini API.
By Lucia Loher · blog.google
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.