Google Teaches Gemini API to Load Agent Tools Only When Needed
The Gemini API has a new feature called Tool Retrieval, now in preview. An agent receives the description of a tool only when the task needs it.
Google has added a feature called Tool Retrieval to the Gemini API. That is the service developers use to build products on Google's Gemini models. The feature is in preview for now. An AI agent is a program that completes tasks by using tools, such as a refund system or a calendar. Normally, the full instructions for every tool are sent with every request. Now the model sees only each tool's name and short description. The full instructions are loaded only when a task needs that tool.
Extra text in a request means more tokens. Tokens are the units of text that developers pay for. Extra text also means a longer wait for an answer. Google says the model also gets worse at picking the right tool as the list grows. The feature has no separate charge. Developers whose agents use dozens of tools stand to gain the most.
The announcement gives no figures on savings. Google itself advises developers to measure the effect on their own traffic. So for now, the benefit is the company's claim, not a measured result. The feature works with Gemini 3.6 and later models. It is not yet clear how much it helps agents with only a few tools.
Source: blog.google
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles