Google Antigravity

Google Antigravity is Google's agentic development platform, designed to help developers build and operate AI agents across the full software development lifecycle. Released as Antigravity 2.0 at Google I/O, it is now generally available and targets everyone from professional enterprise developers to hobbyist vibe-coders. The platform ships as a desktop application, a CLI, an SDK, and a fully-featured IDE, giving teams multiple surfaces to interact with agents.

What It Is

Google Antigravity is an agent-first platform that orchestrates multiple subagents across local environments, enabling complex, long-running workflows to run in the background while other agents handle research, implementation, and testing in parallel. The core abstraction is the Project, which groups conversations, folders, and Git repositories so agents have full cross-repository context. Agents can operate in Local Mode (directly in active folders) or New Worktree Mode (in an isolated Git worktree), and users can steer agent behavior at any point through feedback across every surface and artifact.

The Four Surfaces

Antigravity ships as four distinct but integrated products:

• Antigravity 2.0 — The desktop command center for managing multiple local agents in parallel. Introduces Projects, multi-workspace support, and scheduled messages for automating routine tasks.
• Antigravity CLI — A lightweight, terminal-first interface for running autonomous coding agents, executing shell commands, and managing background subagents entirely from the keyboard.
• Antigravity SDK — A Python-based toolkit for prototyping custom agents with minimal code, iterating on agentic applications, automating software engineering tasks, and running evaluations on top of the Antigravity agent harness.
• Antigravity IDE — A fully-featured agentic IDE with an integrated agent manager, artifacts panel, and deep codebase understanding.

Model Access and Slash Commands

The platform provides access to a range of frontier models including Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet & Opus 4.6, and gpt-oss-120b. Developers interact with agents using slash commands that shape agent behavior:

• /goal — Runs until a task is completely finished without asking for intermediate input.
• /grill-me — Prompts the agent to ask clarifying questions before implementation begins.
• /schedule — Schedules a one-time or recurring instruction via Scheduled Tasks.
• /browser — Explicitly enables browser-in-the-loop behavior, requiring Google Chrome and user permission to start a debugging session.

Audience and Use Cases

The platform is explicitly built for three developer profiles, according to the product site: full-stack developers building production-ready applications with verification tests and artifacts; enterprise developers working in large codebases who need agent-powered scale; and frontend developers who want to automate repetitive UX tasks using browser-in-the-loop agents. The product page also highlights robotics researchers and teams orchestrating multiple parallel agents for rapid feature development.

Update: Antigravity 2.0 at Google I/O

Antigravity 2.0 is the current major release, announced at Google I/O and now generally available. Key additions in this version include the Projects system for grouping conversations and repositories, multi-workspace operation, scheduled messages, and the new Organization plan via Google Cloud — which gives Google Cloud customers access to Antigravity 2.0 and CLI under Google Cloud Terms of Service with consumption-based API pricing and Google Cloud Project integration. The platform currently supports macOS (version 12 Monterey and later, Apple Silicon only — x86 not supported), Windows 10 64-bit, and Linux (glibc ≥ 2.28).