en.Wedoany.com Reported - Google DeepMind has officially launched the Interactions API, advancing Gemini from model invocation to agent execution. This API can simultaneously handle Gemini model calls and agent execution on a unified endpoint. Developers only need to pass a model ID for inference tasks and an agent ID for autonomous tasks.

Previously, after a developer sent a prompt, the API would only invoke the model and return an answer. The key change with the Interactions API is that the API's role has expanded from simple model invocation to managing agent task execution, supporting functions such as agent execution, task state management, tool connectivity, and background processing within a single workflow.

The Managed Agent feature helps developers create a remote Linux sandbox with a single API call. The Gemini agent can reason, execute code, browse the web, and manage files within this sandbox.

VentureBeat believes that the Interactions API is a unified interface designed to make large language models (LLMs) no longer just text generators, but more akin to remote operating systems. This means the model is transitioning from a tool that only generates answers to a system capable of invoking multiple models, executing repeated calls, using tools, and even running backend code.

In terms of cost, the Interactions API stores conversation records and task content on Google servers, eliminating the need for developers to resend the same long context each time. In agent services that frequently use large documents or long conversation records, this can reduce token costs and response latency. Task records for free-tier customers are retained for one day, while paid users can retain them for 55 days. TechCrunch notes that this helps maintain API state and lower prices, but enterprises handling sensitive data must also consider data governance and data residency issues.

Ali Chebik, Group Product Manager at Google DeepMind, emphasized on the official website that the Interactions API was developed based on developer feedback, and this direction will not change after the official release.