Google Makes Interactions API the Default for Gemini Agents
The unified endpoint now leads Google's model-and-agent developer stack, adding managed sandboxes, background jobs and optional server-side state while the older API remains supported.
Google has made the Interactions API its primary interface for building with Gemini models and agents. The updated API brings ordinary model calls, specialised agents, background execution and optional server-managed conversation state behind one endpoint, changing the default integration path for developers using Google AI Studio or the Gemini API.
Google's developer documentation was last updated on August 11, 2026, but it also says the API had been generally available since June. That distinction matters: the current change is not the first public availability of Interactions API. It is Google's decision to make the interface its recommended default, expand its agent-oriented capabilities and direct new projects toward it while keeping the older `generateContent` API supported.
One interface for models and longer-running agents
The central design is an `Interaction`, a resource representing one complete turn in a conversation or task. Its stored record can include user input, model output, reasoning steps and tool calls. A completed interaction can be continued by passing its identifier into a later request, so an application does not have to resend the entire conversation history every time.
Google is also using the interface to narrow the gap between a model request and an agent job. Developers can call a Gemini model directly or specify an agent such as Antigravity or Deep Research. For longer work, background execution lets the server process a task asynchronously instead of requiring the original request to remain open.
Managed Agents extend that model further. Google says one API call can provision a remote Linux sandbox in which an agent can reason, execute code, browse the web and manage files. Built-in tools such as Google Search and Google Maps can be combined with developer-defined functions in the same request. These are meaningful platform additions, but teams still need to test permissions, tool reliability and failure handling in their own applications.
Server-side state adds convenience and data choices
Interactions are stored by default so developers can use server-side continuation, observability and background execution. The current documentation lists a 55-day retention period for paid-tier interactions and one day for free-tier interactions. Paid projects can select shorter deletion windows, and stored interactions can also be deleted programmatically.
Applications that do not want Google to store an interaction can set `store=false`. That choice has functional consequences: stateless requests cannot use the previous-interaction identifier, and the documentation says stateless mode is incompatible with background execution. The migration decision is therefore partly architectural and partly about how an organisation wants conversation state, logs and retention handled.
The older API remains supported
Google now defaults its documentation, Google AI Studio and Gemini API examples to Interactions API, and recommends it for new applications. It expects more long-running model and agent features to arrive there first. Python and JavaScript SDKs already expose the interface.
The company has not announced an immediate shutdown of `generateContent`. Its documentation describes that API as legacy but fully supported, including continued access to mainline Gemini models. Existing teams can therefore migrate deliberately instead of treating this as an emergency cutoff. The practical pressure will come from feature differences rather than a published retirement date.
Those differences also run both ways. Google lists remote MCP support for Gemini 3 as still forthcoming. Video metadata, the Batch API, automatic function calling in Python, explicit caching and custom safety settings are among the capabilities available through `generateContent` but not yet through Interactions API. A team that depends on any of those features should verify parity before moving production traffic.
Why the change matters
The update makes Google's preferred architecture explicit: a stateful interface that can move between direct model inference and more autonomous work without changing the surrounding API pattern. That can simplify applications that mix short answers, tool use and long-running research, while making execution steps easier to inspect.
It also creates a clearer migration checkpoint. Developers no longer have to infer whether Interactions API is an experimental side path or the likely home of future agent features. The remaining question is how quickly its missing capabilities reach parity with the older interface, particularly for teams that require custom safety settings, batch processing or remote MCP.
Status
Confirmed. Google published both the product announcement and current developer documentation. Internal confidence is medium because feature behavior and migration benefits are described by the platform vendor and were not independently tested by AI News Bank.
Sources
- Google — Interactions API becomes the primary interface for Gemini models and agents
- Google AI for Developers — Interactions API overview
Update note: Last reviewed 2026-08-11. We will revise this article if Google changes legacy API support, retention controls or the documented feature gaps.
Sources
- Google — Interactions API announcement — official
- Google AI for Developers — Interactions API overview — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.