🔍 Read the full analysis: How To Use Tool Retrieval In The Gemini API on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google announced Tool Retrieval in preview for the Gemini Interactions API on Nov. 19, 2026. It keeps deferred tools’ names and descriptions visible while loading full parameter schemas only when requested; Google has not published general savings or performance results.
Google announced Tool Retrieval in preview for its Gemini Interactions API on Nov. 19, 2026, as detailed in the original analysis, giving developers a way to load function parameter schemas only when a model requests a particular tool. The change is aimed at agents connected to large tool libraries, where including every schema in every request can use 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 be held back. The model can still see each deferred function’s name and description. When it selects a function, the API retrieves its parameter schema, after which the model can make the function call in the same request.
Google says the interaction steps show the sequence: a tool-search call, a search result containing the schema, and then the function call. The application handles that call and returns its result as usual. Developers can keep frequently used tools loaded and defer less common ones. Google says server-side retrieval requires no index for developers to build or host.
The feature also supports tools exposed through remote MCP servers. Developers using their own tool registry, embedding index or per-user permission filters can instead use client-side retrieval: their application searches its own system and returns matching schemas. That option adds a round trip for each search. Google says the feature works through the Interactions API with Gemini 3.6 and later, including all Gemini 4 models.
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.
How Deferred Schemas Change Requests
Tool Retrieval changes how developers can present large function libraries to an agent. In conventional function calling, a request includes tool definitions and their parameter schemas. With deferred loading, a model initially sees names and descriptions for selected tools, while detailed parameter information is supplied only after it asks for a specific function. That may reduce repeated prompt overhead when a system has many tools but uses only a few for a given task.
Google says the approach is intended to reduce input-token use, latency and tool-selection difficulty as tool libraries grow. Those are design goals, not published results: the announcement provides no benchmark for token savings, response speed or selection accuracy. The practical impact will depend on how often tools are used and how much schema information the model ultimately needs.
Google says there is no separate charge for retrieval, but the search itself adds a small fixed block of input tokens. Each schema that is loaded is billed as input at the model’s standard rate for the rest of the conversation. Savings therefore depend on whether the cost of schemas not loaded outweighs the search overhead and schemas that are retrieved. Google advises developers to measure results on their own traffic.
Products Worth Considering

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

DJI Mini 5 Pro Fly More Combo with DJI RC 2, 1-Inch CMOS

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
From Full Schemas to On-Demand Loading
Function-calling systems give models structured definitions of tools they can use, including the parameters each function accepts. For an agent with a short list of functions, sending those definitions with a request may be straightforward. For a large library, however, supplying every parameter schema can add information that is irrelevant to the current task. Tool Retrieval is Google’s preview approach to making that detailed information available later in the interaction.
Google also says deferred schemas are compatible with prefix caching. Because a deferred schema is added later rather than at the beginning of the prompt, the company says the cached prefix can remain unchanged across turns. That is a product claim in the announcement; no independent evaluation or quantified caching benefit was provided. The source also describes the feature as part of Google’s broader Gemini API tools coverage and points developers to setup documentation.
“Tool Retrieval lets your agent keep its full tool library on standby and load parameter schemas only when the current task needs them.”
— Google, in the announcement quoted by ThorstenMeyerAI.com
Savings and Preview Limits
Google has not supplied a general savings figure, benchmark results or independent evaluations of the feature. The announcement does not quantify changes in total token use, latency or tool-selection accuracy, and it does not specify how results vary across different library sizes or usage patterns. A workload that regularly uses most of its tools may see less benefit than one that rarely needs many deferred schemas.
The feature remains in preview, and Google has not stated a general-availability date or detailed the preview’s limitations. It is also not yet clear from the announcement how accurately tool search performs as libraries grow. Developers should treat the stated benefits as intended outcomes and test them against their own requests and costs.
Testing Against Real Workloads
Developers can consult Google’s setup documentation and try Tool Retrieval with supported models through the Interactions API. Google has invited users to discuss the feature and provide feedback. For teams considering adoption, the next step is to compare requests with and without deferred schemas using representative traffic.
Those tests can track the fixed search-token overhead, how many schemas are retrieved, total input-token use and request behavior. Teams using client-side retrieval should also account for its additional round trip. Google has not announced a date for moving beyond preview, so availability and any later changes remain subject to future updates.
Key Questions
What is Tool Retrieval in the Gemini API?
It is a preview feature for the Gemini Interactions API that lets developers defer full function parameter schemas and load them when a model requests a tool. Names and descriptions remain available before the full schema is retrieved.
How do developers enable server-side Tool Retrieval?
Google’s described setup uses a tool_search tool and the defer_loading: true setting on functions whose schemas should be held back. The API then retrieves a schema when the model requests that function.
Does Tool Retrieval reduce Gemini API costs?
It may reduce input tokens when schemas are not loaded, but Google has not published a general savings figure. The search adds a small fixed block of input tokens, and retrieved schemas are billed at the model’s standard input rate, so developers need to measure their own workloads.
Which models support the feature?
Google’s announcement says Tool Retrieval works through the Interactions API with Gemini 3.6 and later, including all Gemini 4 models.
Is Tool Retrieval generally available?
No. Google announced it as a preview on Nov. 19, 2026. The announcement did not provide a general-availability date or specify all preview limitations.
Primary source: Google AI · via ThorstenMeyerAI.com
