Google previews Gemini API Tool Retrieval for large tool libraries
First merged summary.
Google is previewing Gemini API Tool Retrieval so agents load tool schemas on demand instead of sending them all up front.
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, 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. The feature is available on Gemini 3.6 and later, including Gemini 4; a sample request uses gemini-4-hydrogen. There is no separate fee; tokens are billed at the model's standard rates.
- Preview feature in the Gemini Interactions API, not a general release.
- Supported on Gemini 3.6 and later, including Gemini 4.
- Deferred schemas are loaded only after the model requests a tool by name.
- Search can be server-side, client-side, or against remote MCP servers.
- No separate fee; usage is billed at standard model token rates.
- Prefix caching stays stable because deferred schemas arrive later in the conversation.
What each outlet adds
beyond the shared facts- Google AIAPI mechanism, model support, billing, and caching behavior.
- Vendors
- Products
- Gemini APIInteractions API
- Organizations
- Google DeepMind
- AI models
- Gemini 3.6Gemini 4gemini-4-hydrogen
Coverage
oldest first · one row per article · UTCGoogle previews Tool Retrieval in the Gemini Interactions API
Agents can defer tool schemas and load them on demand, cutting input tokens on Gemini 3.6 and later, including Gemini 4.
Google previews tool retrieval in the Gemini API
The preview is aimed at applications that call large tool libraries through the Gemini API.