Google Interactions API goes GA for Gemini agents

Google has made its Interactions API generally available, setting it up as one interface for Gemini models, tools and agent workflows. In a company announcement, Google said the API is now its primary way to work with Gemini and agents, replacing the older pattern of stitching together separate model calls, tool calls and state handling.

The release takes a product first opened in beta in December 2025 into Google’s recommended production path. Developers can keep conversation state on the server, combine models and tools inside one workflow, and hand long-running jobs to background execution, Google said. That is routine-sounding plumbing, but it is often what turns an agent demo into an engineering project: queues, retries, session stores and recovery logic all have to work once a system does more than answer a single prompt.

“the Interactions API has reached general availability and is now our primary API for interacting with Gemini models and agents”

• Google, Interactions API announcement

In its developer documentation, Google called the Interactions API the “most straightforward way” to build with Gemini models and agents. The line matters because the contest among Google, OpenAI, Anthropic and Microsoft has moved beyond raw model quality into the software layer around agents. A single endpoint that keeps state, calls tools and resumes work later is easier to sell to an internal development team than a bundle of APIs and orchestration code.

For local software teams, the pitch is practical. An Australian product group building a support bot, coding assistant or internal search tool can keep state and long-running jobs on Google’s side instead of running its own session store or leaving a browser tab open while a task completes. Smaller teams get agent features without standing up another layer of cloud infrastructure first. For a Sydney SaaS team shipping overnight updates to Asia-Pacific customers, that can matter more than another benchmark gain.

The easier setup comes with a trade-off: more of the agent runtime sits inside Google’s hosted stack.

Google said previous interactions on paid tiers can be retained for 55 days, a detail that points to the platform choice behind the simpler setup. Google also said its Flex tier can cut costs by as much as 50 per cent for latency-tolerant work, aiming the offer at batch jobs such as document processing, code analysis and overnight summarisation rather than interactive consumer chat.

Outside Google’s own materials, API Evangelist argued that making the Interactions API the “front door” to Gemini is as much a platform move as a product release. One default surface gives Google a place to steer how developers use models, tools and state across future services.

For Google, the general-availability label is less about a new model than about lowering friction around the ones it already sells. For developers choosing an agent stack in 2026, the test is whether one endpoint saves engineering time, or simply trades home-built complexity for deeper reliance on one vendor.