Google’s Interactions API has reached general availability and is now Google’s primary interface for building with Gemini models and agents. It replaces the legacy generateContent endpoint as the default for all new documentation and tooling 1.
The GA launch ships a stable, production-ready schema alongside new agent-focused capabilities built natively for stateful agentic workflows. These capabilities are designed for multi-step tasks, a design shift from the single-turn, short-context inference use cases the legacy generateContent API was built to handle 1.
Key New Features in the Interactions API GA
The headline addition for the GA release is Managed Agents: a single API call provisions a fully managed remote Linux sandbox for agent workloads. Each sandbox supports full agentic operations, including reasoning over task requirements, executing arbitrary code, accessing live web content, and reading or writing files within the isolated environment 1.
Google’s in-house Antigravity coding agent is pre-installed as the default out-of-the-box option for these sandboxes. Developers can also deploy custom agents defined via user-provided instructions, custom skill libraries, and proprietary data sources 1.
Developers can set a background=True flag on any API call to run interactions asynchronously on Google’s servers. This eliminates the need to maintain persistent connections for tasks that take minutes or hours to complete, reducing operational overhead for long-running agent workflows 1.
Tooling improvements let users mix built-in Google tools (including Search and Maps) with custom function calls in a single request. Tool results now support image outputs alongside text, expanding multimodal use cases for agent workflows 1.
Deep Research capabilities have been upgraded with two distinct agent versions optimized for speed or depth, plus collaborative planning support. The update also adds native chart and infographic generation, and multimodal grounding that accepts images, PDFs, and audio as input 1.
The release previews support for the upcoming Gemini Omni model, alongside native media generation tools. These include image generation via the Nano Banana 2 model with Google Image Search grounding, music generation via Lyria 3, and expressive multi-speaker text-to-speech 1.
The API schema has been simplified from a role-based structure (used in the legacy generateContent API) to a step-based model. Every discrete action (user input, model thought, function call, output) is logged as a typed step for granular observability, a design choice that prioritizes debugging for multi-step agent tasks over backward compatibility with existing single-turn chat patterns 1.
Cost tiers include a Flex option for latency-tolerant workloads and a Priority tier for low-latency use cases. Paid tier users get 55-day retention of past interaction logs for audit and debugging purposes 1.
How Does the Interactions API Compare to Legacy generateContent?
Google stated the legacy generateContent API will stay fully operational with ongoing support, and will be updated to include new mainline Gemini model releases for the foreseeable future. The company noted all frontier capabilities for long-running models and agent workflows will ship exclusively on the Interactions API going forward 1.
This split is due to the Interactions API’s ground-up design for stateful, multi-step agentic tasks, while the legacy endpoint was built primarily for single-turn or short-context chat use cases 1.
A published migration guide maps every field from the legacy generateContent schema to the new step-based Interactions API format. All official documentation now defaults to Interactions API code snippets, with a toggle available to switch examples back to the legacy format 1.
Google recommends all new projects use the Interactions API, and expects most existing agent-focused workloads to migrate over time to access new features. No end-of-life date has been announced for the legacy generateContent API as of the June 22, 2026 GA launch 1.
What Ecosystem Support Exists for the Interactions API?
The Interactions API is available immediately via Google’s official Python and JavaScript SDKs. Third-party integrations are already live for supported partners including LiteLLM, Eigent, and Agno 1.
For developers building with coding agents, Google released a dedicated gemini-interactions-api skill. This skill injects best-practice patterns for the API (including streaming, function calling, structured output, and Deep Research usage) directly into an agent’s context to reduce implementation friction 1.
API keys are available via Google AI Studio, and developers can access full API reference documentation and the migration guide through the official Gemini developer portal. Google also noted it will continue to prioritize developer feedback for the Interactions API, with a public developer forum open for feature requests and bug reports 1.
Bottom line: New Gemini projects should default to the Interactions API immediately to access the stable step-based schema, managed Linux sandbox agent tools, and 55-day paid-tier log retention, while teams running existing long-running agent workflows should prioritize migration to avoid missing exclusive frontier features that will not launch on the legacy generateContent endpoint.
