cd /news/ai-agents/google-adds-tool-retrieval-to-the-ge… · home › topics › ai-agents › article
[ARTICLE · art-172737] src=cryptobriefing.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Google adds Tool Retrieval to the Gemini API to tame tool-heavy AI agents

Google launched Tool Retrieval in the Gemini API as a preview feature inside the Interactions API, letting agents fetch only the tool parameter schemas they need instead of sending every schema with each request. Google says the old full-schema approach inflates input tokens, adds latency, and lowers tool-selection accuracy as the candidate list grows. The Interactions API reached general availability in mid-2026, and Google rolled out multimodal enhancements to File Search in May 2026; pricing and general availability for Tool Retrieval remain unannounced.

key takeaway

Tool Retrieval in the Gemini API's Interactions API fetches only the relevant tool schemas per call instead of shipping every schema, cutting input tokens and latency and improving tool-selection accuracy for agents wired to dozens or hundreds of tools.

by read3 min views1 publishedNov 19, 2026
Google adds Tool Retrieval to the Gemini API to tame tool-heavy AI agents
Image: Cryptobriefing (auto-discovered)

gemini icon

The preview feature lets agents pull only the tool schemas they need instead of shipping the full list with every request

Google has launched Tool Retrieval in the Gemini API, a preview feature built for AI agents that juggle a lot of tools at once.

It lives inside the Interactions API. It targets a problem that grows with every tool a developer adds: agents get slower, cost more to run, and get worse at choosing the right tool.

What Tool Retrieval actually fixes #

Start with how agents use tools. An AI agent can call outside functions, such as a search engine, a file lookup, or a custom piece of business logic.

To pick the right one, the model needs to know what each tool does and what inputs it expects. That description is called a parameter schema, and it is basically the tool’s instruction manual.

Until now, an agent wired to dozens or even hundreds of tools would send every one of those full schemas along with every request. Google says this approach piles up quickly.

Google points to three specific costs of the old approach:

  • Inflated input tokens. Every schema counts toward what the model reads, and models bill by the token.
  • Added latency. More text to process means a slower response.
  • Lower tool-selection accuracy. As the candidate list gets longer, the model gets worse at picking the correct tool.

Tool Retrieval changes the default. Rather than shipping the full parameter details for every tool on every call, the system retrieves only the relevant schemas when they are needed.

AI, tech, and the markets they move—in one daily briefing.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

Where it fits in Google’s developer stack #

Tool Retrieval is part of a broader push by Google to make tool use a first-class feature of the Gemini API.

The API already ships with built-in tools, including Google Search, File Search, and Code Execution. Developers can also bring their own functions through custom function calling.

The Interactions API is the home base for all of it. It reached general availability in mid-2026 and is the preferred interface for new applications built on Gemini.

Its design leans into multi-turn workflows and server-side state management. Translated: the API is built for conversations and tasks that unfold over many steps, with Google keeping track of the running context on its side rather than forcing developers to resend everything each time.

The feature also follows other recent upgrades. In May 2026, Google rolled out multimodal enhancements to File Search, part of the same steady stream of improvements aimed at developers building agent-driven applications.

What this means for developers building agents #

The clearest impact lands on teams running large, tool-heavy agents. Those are the setups where full-schema requests hurt most, since the overhead scales with the number of tools connected.

There are caveats worth keeping in mind. The feature is in preview, which generally means behavior, availability, and interfaces may change before a stable release.

Developers will also want to see how retrieval performs in practice. Any system that decides which tools to surface introduces its own failure mode: if the retrieval step misses the right tool, the model never gets a chance to pick it.

Things to watch from here include how Google prices the feature and whether it moves from preview to general availability.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

── more in #ai-agents 4 stories · sorted by recency
── more on @google 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/google-adds-tool-ret…] indexed:0 read:3min 2026-11-19 · —