Interactions API: The Key Interface for Gemini Models and Agents

Google has announced the general availability (GA) of the Interactions API, designating it as the default API for interacting with Gemini models and agents. Having started public beta in December 2025, this API now has a stable schema and includes powerful features reflecting developer needs.

Key updates include the following:

• Managed Agents: Provision a remote Linux sandbox with a single API call to create agents capable of code execution, web browsing, and file management.
• Background Execution: Long-running tasks can be processed asynchronously via the background=True setting.
• Tool Improvements: Built-in tools such as Google Search and Google Maps can be mixed with user-defined functions in a single request, and results can now return images.
• Deep Research Upgrade: Offers two agent versions to choose between speed and depth, collaborative planning, native charts, and multimodal grounding capabilities.
• Media Generation: Supports image generation via Nano Banana 2, music generation via Lyria 3, and expressive voice generation via multi-speaker TTS.

The Interactions API has been set as the default interface for Google AI Studio and the Gemini API. The existing generateContent API will continue to be supported, but stateful, agent-centric workflows and the latest features will be centered around the Interactions API going forward.