What changed
Google has moved the Gemini platform toward a single default interface for both model calls and agent workflows. In its June 22 announcement, Google said the Interactions API is now generally available, has a stable schema, and is now the primary API for interacting with Gemini models and agents. The companion docs now recommend it for all new projects.
The practical shift is bigger than a normal API graduation. Instead of treating chat completions, tool use, long-running jobs, and agent execution as separate patterns, Google is consolidating them behind one interaction model.
Why builders should care
For portfolio projects, internal tools, and production apps, the biggest win is simpler architecture. The Interactions API gives developers one interface for plain model inference, managed agents, and built-in tools like Search and Maps. Google is also positioning it as the place where new models and new agentic capabilities will launch first, which makes it the safer default for greenfield Gemini work.
The docs also highlight a cost and latency angle: server-side state lets you continue conversations with previous_interaction_id instead of resending the full transcript every turn, which improves cache hit rates and can reduce token spend for multi-turn apps.
The notable new capabilities
- Server-side state: continue a conversation by passing
previous_interaction_id. - Background execution: set
background=truefor long-running tasks. - Managed Agents: provision a remote Linux sandbox for agents that can reason, execute code, browse the web, and manage files.
- Tool composition: combine built-in Google tools with your own functions in a single request.
- Unified model and agent access: call standard Gemini models and agent products like Deep Research through the same API family.
The migration signal
Google is explicit that the older generateContent API is still supported, but it is now considered legacy in the Interactions docs. That means existing integrations do not need an emergency rewrite, but new work should start on Interactions unless there is a specific compatibility reason not to.
There are still tradeoffs to evaluate. The docs say storage is enabled by default, with retained interactions available for server-side state and background execution. Paid-tier interactions are retained for 55 days and free-tier interactions for 1 day unless developers opt out with store=false, which also disables some of the convenience features.
Bottom line
This is the clearest sign yet that Google wants Gemini app development to converge on one API surface. If you are building a new agent, research workflow, or multimodal assistant, the Interactions API is now the default path rather than an advanced option.