Thursday, November 19, 2026
BP·InfoAI Briefing

The AI news that matters, explained for business and IT professionals.

Tools & Products

Google previews Tool Retrieval in the Gemini API so agents can load tool definitions on demand

Google says Tool Retrieval is now in preview in the Gemini Interactions API. Developers can mark tools as deferred, so the model sees only names and descriptions until it needs one, and the API then loads the full schema within the same request. Google says there is no separate charge, and it gives no benchmark figures for token savings.

Why it matters

Agents connected to many tools can burn a large share of their context window on tool definitions. Deferred loading, similar to approaches from other vendors, is becoming a standard pattern for large tool libraries. It works with Gemini 3.6 and later models, including the Gemini 4 family, and a client-side mode lets teams run their own lookup.

For business & IT

Measure token use and tool-selection accuracy on your own traffic before and after enabling it, since Google does not publish savings figures.

#gemini-api#agents#tool-use#google

More in Tools & Products

See all →

Google’s Gemini app for macOS gains computer use for multi-step tasks

Google says the Gemini app on macOS can now use the computer to complete multi-step tasks across local files, desktop apps and the web, according to The Verge. Google suggests uses such as filling in forms, making presentations and listing items for sale. TechCrunch also covered the launch.

The Verge

Google rolls out Gemini 4 Neon, a mid-tier model priced at $3 per million input tokens

Google DeepMind says Gemini 4 Neon is rolling out in the Gemini API, Google AI Studio, Gemini for business, Android Studio and, for paid subscribers, the Gemini app. It sits between Carbon and Hydrogen in the Gemini 4 family and costs $3 per million input tokens and $15 per million output tokens, with a 1-million-token context window. The company says it costs 32% less per task than Carbon at its default setting.

Google DeepMind