🔍 Read the full analysis: Meet Tool Retrieval For The Gemini API on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Google announced Tool Retrieval in preview for the Gemini Interactions API on Nov. 19, 2026. It lets developers defer full function parameter schemas until a model requests a tool, but Google has not published general savings figures, and costs depend on which schemas are loaded.
Google announced Tool Retrieval in preview for the Gemini Interactions API on Nov. 19, 2026, giving developers a way to load full function parameter schemas only when a model requests a tool, as explained in the original analysis. The change is aimed at agents with large tool libraries, where sending every schema with each request can add input tokens even when most tools are not needed.
To use the server-side feature, developers add a tool_search tool and set defer_loading: true for functions whose full schemas should not be included up front. The model can still see each deferred tool’s name and description, but it does not receive the parameter schema until it requests that tool by name. The API then supplies the schema so the model can make a function call in the same request.
Google says the interaction’s steps show the sequence: a tool search call, a result containing the requested schema, and a function call. The application handles that function and returns its result as usual. Developers can keep frequently used tools loaded from the start and defer less common ones. Google says server-side retrieval requires no index for developers to build or host, and the feature can also retrieve tools exposed through remote MCP servers.
Google also describes a client-side retrieval option for teams using their own registry, embedding index or per-user permissions filter. The application searches its own system and returns matching schemas, but this approach adds a round trip for each search. The announcement says Tool Retrieval works through the Interactions API with Gemini 3.6 and later, including all Gemini 4 models. Google says there is no separate charge for retrieval; the search uses a small fixed block of input tokens, and loaded schemas are billed as input at the model’s standard rate for the rest of the conversation.
Introducing Tool Retrieval in the Gemini API
Google’s preview feature loads tool parameter schemas on demand, helping agents work with large tool libraries while keeping unused details out of the initial request.
What changes for agents
Tool names and descriptions remain visible; full parameter schemas load when requested.
Developers mark selected functions for deferred loading.
The model requests a tool by name and receives its schema.
The model calls it, and the application handles the result as usual.
Choose a retrieval approach
Server-side retrieval suits managed tool libraries; client-side retrieval supports custom search systems.
- Uses the tool_search tool.
- No index to build or host.
- Supports remote MCP server tools.
- Searches a custom registry or index.
- Can apply per-user permission filters.
- Adds a round trip for each search.
Understand the cost trade-off
Tool Retrieval has no separate charge; standard model token rates apply.
Skip unused schemas
Savings depend on whether schemas left unloaded outweigh schemas retrieved.
Search and loaded schemas
Search adds a small fixed block of input tokens; loaded schemas count as input for the rest of the conversation.
Managing Large Gemini Tool Libraries
Tool Retrieval addresses a scaling problem for agents connected to many functions: each request may otherwise carry detailed schemas for tools irrelevant to the current task. Keeping those details out until needed could reduce repeated prompt overhead and leave more of the input context available for the user’s request and other information. It could also make tool selection easier when a model has a large set of functions to choose from, though Google has not supplied evidence quantifying that effect.
The financial and performance outcome depends on actual use. Retrieval adds input tokens, and every schema that is loaded is billed at the standard input rate. A system that regularly uses most of its tools may see less benefit than one with many rarely used functions. Google says savings depend on which schemas are loaded and recommends measuring results against a team’s own traffic. It has not provided a general savings figure or benchmark for token use, latency or selection accuracy.
Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

Amazon Kindle Scribe Colorsoft 64GB with Premium Pen

DJI Mini 5 Pro Fly More Combo with DJI RC 2, 1-Inch CMOS
As an affiliate, we earn on qualifying purchases.
From Preloaded Schemas to On-Demand
In conventional function calling, developers send tool definitions, including parameter schemas, with a request. Google’s preview changes when those details reach the model: names and descriptions remain available, while full schemas for selected functions are supplied only after a tool search. The function still runs through the application’s usual handling and response process.
Google says deferred schemas are compatible with prefix caching. Because the schemas are added later rather than included in the initial prompt, the company says the cached prefix can remain unchanged across turns. Tool Retrieval is presented as one option for larger tool libraries, rather than a replacement for keeping frequently used tools available from the beginning.
““Tool Retrieval lets your agent keep its full tool library on standby and load parameter schemas only when the current task needs them.””
Savings and Preview Limits
The announcement does not report benchmark results for token savings, latency or tool-selection accuracy, and it gives no general-availability date. Google says the cost effect varies with how many schemas are loaded and advises developers to measure their own traffic. The available material does not specify all preview limitations or provide independent evaluations of model performance as tool libraries grow.
It is also unclear how the feature will perform across different library sizes and usage patterns. Server-side retrieval avoids the need for a developer-hosted index, while client-side retrieval offers control over searching and permissions at the cost of an additional round trip. Developers will need to compare the approaches in their own systems to determine whether the added search step is worthwhile.
Testing Before Broader Availability
Developers can consult Google’s setup documentation and test Tool Retrieval with supported models through the Interactions API while the feature is in preview. Google has invited users to discuss the feature and provide feedback. Teams evaluating it can compare requests with and without deferred schemas, tracking total input tokens, latency, tool-selection behavior and the effect of search steps under representative workloads.
Google has not announced when the preview will end or when the feature might become generally available. Until it publishes more measurements or availability details, the practical next step is workload-specific testing rather than assuming the feature will reduce costs or delays.
Key Questions
What is Gemini Tool Retrieval?
Tool Retrieval is a preview feature for the Gemini Interactions API. It lets developers defer full function parameter schemas and load them when the model requests a tool.
Can the model see a deferred tool before its schema loads?
Yes. Google says the model can see the tool’s name and description, but not its full parameter schema. The API supplies that schema after the model requests the tool by name.
Does Tool Retrieval guarantee lower costs or faster responses?
No. Google has not provided general savings or latency figures. Retrieval adds a small fixed block of input tokens, and loaded schemas are billed at the model’s standard input rate, so results depend on each workload.
Which Gemini models support the preview?
Google says the feature works through the Interactions API with Gemini 3.6 and later, including all Gemini 4 models.
When will Tool Retrieval be generally available?
Google has described Tool Retrieval as a preview and has not announced a general-availability date.
Primary source: Google AI · via ThorstenMeyerAI.com
