🔍 Read the full analysis: A Guide To Tool Retrieval For Gemini API Developers 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 keep tool names and descriptions visible while loading full parameter schemas only when requested; Google has not published general savings or performance figures.
Google announced Tool Retrieval in preview for its Gemini Interactions API on Nov. 19, 2026, giving developers a way to load full function parameter schemas only when a model requests a tool. The feature is intended for agents connected to large tool libraries, where including every schema in each request can add input tokens; Google has not published a general estimate of potential savings.
With the server-side option, developers add a tool_search tool and set defer_loading: true for functions whose schemas should be held back. The model can still see each deferred function’s name and description, but not its parameter schema. If it requests a function by name, the API returns the schema so the model can make the function call within the same request. The application then executes the function and returns its result as usual.
Google says the interaction steps show the tool search call, the result containing the schema and the subsequent function call. Developers can keep common tools loaded from the start and defer less frequently used ones. The company says server-side retrieval requires no index for developers to build or host. It also supports tools exposed by remote MCP servers when deferred loading is enabled.
Teams with 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 approach adds a round trip for each search. Google says Tool Retrieval has no separate feature charge, but the search uses a small fixed block of input tokens, and loaded schemas count as input at the model’s standard rates 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 developers building agents with dozens or hundreds of functions. When a request includes a full schema for every available tool, parameter details for functions that are not used can occupy prompt space. Deferring those details may reduce that overhead and leave more room for the task and conversation.
The practical result depends on how an agent uses its tools. Google says savings occur when schemas that are never loaded outweigh the schemas and search tokens that are added. A system that uses most tools frequently may see little reduction, while client-side retrieval also adds a network or application step. The feature is therefore a way to change how tool definitions enter requests, not a guaranteed reduction in cost, latency or errors.
For developers, the interaction trace offers a way to inspect which tool was searched for and when its schema was supplied. That can help teams test behavior alongside token use. But Google has not provided independent evaluations or benchmark figures showing how retrieval affects selection accuracy or performance across different tool libraries.
Products Worth Considering

Apple MacBook Air 13-inch with M5 Chip, 16GB Memory, 512GB SSD, Liquid Retina Display, and 12MP Center Stage Camera – Midnight

Jackery Solar Generator 1000 V2 with 100W Solar Panel

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black
As an affiliate, we earn on qualifying purchases.
From Full Schemas to On-Demand Loading
In conventional function calling, the request includes tool definitions and their parameter schemas. Google’s preview changes that arrangement for deferred functions: the model initially receives names and descriptions, then the API supplies a schema after the model requests the relevant tool. The function itself still runs through the application’s usual execution and result-return process.
Google says the feature works with Gemini 3.6 and later, including all Gemini 4 models, through the Interactions API. It also says deferred schemas are compatible with prefix caching: because the schemas are added later rather than at the start of the prompt, Google says the cached prefix can remain the same across turns. These are company statements in the announcement, not published independent test results.
“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 Remain Unknown
Google has not supplied a general figure for token savings, latency changes or tool-selection accuracy, nor does the announcement include benchmark results. The cost and performance effects will depend on library size, tool-use frequency, request design and, for client-side retrieval, the added search round trip. Google advises developers to measure results against their own traffic.
The feature is in preview, and Google has not specified a general-availability date or detailed preview limitations. The announcement also does not establish whether tool retrieval maintains selection accuracy as libraries grow. Those points remain open pending further documentation, testing and independent evaluation.
Testing During the Gemini Preview
Developers can consult Google’s setup documentation and test Tool Retrieval with supported models in the Interactions API. Google has invited users to discuss the feature and share feedback. For teams considering adoption, the next step is to compare requests with and without deferred schemas using representative workloads, tracking total input tokens, latency and tool-selection behavior.
Further updates may clarify preview limitations and whether Google plans a general release. Until then, the announcement establishes the available mechanism and supported model range, but not a typical return on using it. Any decision to deploy should rest on the team’s own measurements and requirements.
Key Questions
What does Gemini Tool Retrieval do?
It lets developers defer a function’s full parameter schema until the model requests that tool. The model can initially see the tool’s name and description.
Does Tool Retrieval guarantee lower API costs?
No. Google says there is no separate retrieval charge, but search and loaded schemas use input tokens at standard rates. Savings depend on which schemas are loaded and should be measured against actual usage.
Can developers use their own tool search system?
Yes. Google describes a client-side retrieval option for applications using their own registry, index or permission filters. It requires the application to return matching schemas and adds a round trip for each search.
Which Gemini models support the preview?
Google says Tool Retrieval 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 announced the feature in preview, but has not given a general-availability date.
Primary source: Google AI · via ThorstenMeyerAI.com
