Google Cloud Launches Agent Platform With Seven-Day Runtime

Google Cloud launched the Gemini Enterprise Agent Platform on April 22, 2026, giving businesses tools to build and manage AI agents that can maintain context and execute multistep workflows for as long as seven days.

The platform, introduced at Google Cloud Next in Las Vegas, expands the company’s enterprise AI offering beyond one-off conversational responses. It combines agent development, managed runtime, memory, identity, governance and model access in a single system.

Agent Runtime supports long-running operations for up to seven days. It allows agents to preserve state while working across extended tasks, while Memory Bank generates, stores and retrieves long-term context across multiple sessions.

The platform provides access to more than 200 foundation models through Model Garden, including Claude models. Its main components include the Agent Development Kit, Agent Studio, Agent Runtime, Memory Bank, Agent Identity, Agent Registry and Agent Gateway.

“Enterprises don't just need AI to answer questions or complete isolated tasks,” Maryam Gholami, senior director of product for Gemini Enterprise, said.

Gemini Enterprise also includes an Inbox for monitoring long-running agents. The system can notify users when an agent needs input, encounters an error or completes its work. Integrations include Google Workspace, Microsoft 365, partner agents and data connectors.

Agent Identity gives agents a distinct identity for use with Agent Runtime. Agent Identity for Agent Runtime was generally available, while Agent Identity for the Gemini Enterprise Agent Platform was in preview as of May 6, 2026.

The platform is designed to help companies build, deploy, monitor and govern agents across internal systems and partner services. Its approach combines persistent execution with tools for managing state, identity and access across enterprise workflows.

“Agent Platform enables this innovation to scale responsibly by combining deterministic business rules with probabilistic reasoning,” Matt Olson, chief innovation officer at Burns & McDonnell, said.