Google rolls out the Interactions API, its new primary path to Gemini

Google's new Interactions API is now the primary path to Gemini models and agents, replacing generateContent with agentic workflows and managed sandboxes.

Google's Interactions API is now available to everyone and becomes the primary way to interact with Gemini models and agents. Opened in public beta last year, it is designed from the outset for agentic workflows and server-side state management, whereas the old generateContent API, still supported, primarily aimed at model invocation.

The principle lies in a single entry point: passing a model ID for inference, or an agent ID for an autonomous task, and switching to background processing via a simple parameter for long-running tasks. Among the new features, Managed Agents provision in a single call a remote Linux sandbox where the agent reasons, executes code, navigates, and manages files, with the Antigravity agent serving as the default, and the possibility to define custom ones equipped with instructions, skills, and sources. The schema has been simplified, each action, input, thought, tool call, or output, becoming a typed step rather than a role. Tools can be mixed in a single request, combining Google search and custom functions, their results capable of returning images. Additionally, an enhanced Deep Research and the generation of images, music, and multi-speaker voices are included.

Regarding costs, Flex and Priority tiers arbitrate between expenditure and latency, with Flex reducing the cost by half, and interactions are retained for fifty-five days on the paid offering. Google embraces an ecosystem designed for agents, with a dedicated skill injecting its best practices into the context of a code agent, and warns that cutting-edge capabilities for long-running models and agents will increasingly land on this API alone.