Tool Retrieval in the Gemini API (Preview)
Lets developers give an agent hundreds of tools without sending every tool definition on each request.
Details
- Tools are marked with defer_loading, and the API searches the tool library and loads a function's schema only when it is needed
- Retrieval runs through a tool_search tool or client-side execution, and remote MCP servers also support deferred loading
- Cuts token use and keeps the prefix cache intact
- Works with Gemini 3.6 and later models, including all Gemini 4 models, on the Interactions API
Best for
Agents that need a large library of tools without paying for all of them in every prompt
Use it when
Your tool list has grown big enough that the model picks the wrong tool or the prompt gets expensive
Who benefits
Developers building agents on the Gemini API
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →