Google has introduced Tool Retrieval in the Gemini API, now available as a preview feature within the Interactions API. This new capability addresses challenges faced by agents that connect to numerous tools, where sending full parameter schemas for every tool on each request can lead to inefficiencies.
Previously, agents with dozens or hundreds of tools would send all tool parameter schemas with every request. This practice inflated input token counts, increased latency, and decreased the accuracy of tool selection as the list of candidates grew. Tool Retrieval aims to mitigate these issues by optimizing how tool information is handled.
To activate Tool Retrieval, developers add a 'tool_search' tool to their request and set 'defer_loading: true' on specific tools. The model still receives the name and description of all deferred tools, allowing it to understand available functionalities, but their parameter schemas remain out of context. When a tool is required, the model requests it by name, the API loads the full schema, and the model then executes the function call, all within a single API request.
The implementation requires minimal code changes: adding the new 'tool_search' tool and a flag to defer loading on desired tools. Gemini handles the search process on the server side, eliminating the need for developers to build or host an index. The entire process, from tool search to function call, is logged in the interaction steps, aiding in debugging tool names and descriptions.
This feature is significant for developers building agents that interact with a large number of external tools. By reducing token usage and latency, it allows for more complex and responsive agent behaviors without incurring prohibitive costs or performance penalties. It also contributes to more precise tool selection by the model, as it only processes relevant schema information when necessary.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's briefNew every morning, and the back catalogue is archived by date.
Google has launched Tool Retrieval in the Gemini API, available in preview, which allows agents to manage large tool libraries more efficiently. This feature defers loading of tool parameter schemas until needed, reducing input token counts, latency, and improving tool-selection accuracy.