Google has introduced Antigravity, a new Gemini 3–based software development platform that allows developers to build integrated agent workflows within a single application.

Antigravity is designed as a free application that combines coding, testing, and browser actions through AI agents capable of operating autonomously inside the editor, terminal, and browser. Developers can supervise every step through detailed logs, workflow plans, and screenshots that document the agents’ actions.

The platform is available on Windows, macOS, and Linux. In addition to supporting Google’s own models such as Gemini 3 Pro, Antigravity is compatible with external models, including Claude Sonnet 4.5. Its editor interface resembles a traditional IDE, while the manager view enables users to run multiple agents across different workspaces.

Google says Antigravity aims to push software development toward a task-driven workflow, where humans set direction and AI agents execute the work. “We want to make software building more efficient by letting agents handle the execution details,” the company said in an official statement.

Released as a public preview, Antigravity offers what Google calls “generous limits” for Gemini 3 Pro usage. The launch highlights the rapid acceleration of agent-based tools into everyday programming practices.

Google stated that Antigravity will continue to evolve throughout the preview period, focusing on improving agent stability and expanding model integrations in future updates.