Google Introduces Tool Retrieval in Gemini API
Google has launched Tool Retrieval in preview within the Gemini Interactions API, allowing developers to load tool parameter schemas on demand rather than sending the entire tool library with each request. This feature aims to reduce input tokens, improve latency, and enhance tool-selection accuracy as the number of tools increases.
1 min read1 source
Key takeaways
30-second read- 1Developers can optimize API usage by reducing unnecessary token counts, leading to potential cost savings when using the API.
- 2The ability to load tool schemas on demand allows for more efficient handling of large tool libraries, improving application performance.
- 3This feature simplifies the integration of new tools mid-conversation, enhancing user experience and flexibility in tool usage.
Confirmed
Official source: Google AI Blog
Reported
No secondary reporting linked
Verify yourself
Eligibility, local prices and dates for your car, state or utility

Google has introduced Tool Retrieval in the Gemini Interactions API, currently available in preview. This new feature allows developers to load tool parameter schemas only when needed, rather than sending the entire tool library with each prompt. This approach is designed to reduce input token counts, decrease latency, and improve tool-selection accuracy, especially when an agent is connected to numerous tools.
To implement Tool Retrieval, developers need to add a tool_search tool to their requests and set defer_loading: true for the tools they wish to hold back. This means that while the model can see the names and descriptions of all deferred tools, the detailed parameter schemas are not included in the initial context. When a tool is required, the model requests it by name, prompting the API to load the full schema within the same request. This streamlines the process, allowing for efficient function calling and debugging.
Additionally, Tool Retrieval integrates well with prefix caching, ensuring that deferred schemas do not clutter the start of the prompt, thus maintaining a consistent cached prefix throughout the conversation. It also supports remote MCP servers, enabling developers to connect large servers without needing to list every tool individually.
For those who maintain their own tool registries or embedding indexes, client-side retrieval is also an option. This allows for dynamic tool exposure mid-conversation, enhancing flexibility in tool management. Importantly, Tool Retrieval does not incur additional charges; it operates under the standard billing rates for input and output tokens, making it a cost-effective solution for developers managing extensive tool libraries.
Sources
1 official- [1]Introducing Tool Retrieval in the Gemini APIOfficial
Google AI Blog · blog.google · published Nov 19
Newsletter
Weekly AI signal. No spam.




Comments
Sign in to join the discussion
Loading comments…