Feature lists are easy to write and hard to use. This page explains what each part of Antigravity does in practice, when it helps, and where it does not.
A control room rather than a chat log. Spawn an agent per workspace, give each one a task, and watch the queue rather than a scrolling transcript.
The agent launches your dev server, opens the page and drives it. It clicks, types, waits for the network and captures what it sees at each step.
Plans, task lists, command logs, screenshots and a closing walkthrough — generated as the work happens, not written up afterwards.
VS Code lineage: your keybindings, your themes, most of your extensions. Plus tab completion and an embedded agent that can act on the file you are in.
Commands run in a bounded shell with output captured. Risky operations stay visible and auditable instead of vanishing into a background process.
Correcting an agent by retyping the whole request is wasteful. Antigravity lets you comment on the sentence in the plan or the region of the screenshot that is wrong.
Corrections and conventions are written back and retrieved on later runs, which is the difference between an assistant that learns your codebase and one that starts fresh daily.
| Platform | Minimum | Notes |
|---|---|---|
| Windows | Windows 10, 64-bit or later | ARM64 devices should check the release notes first |
| macOS | macOS 12 Monterey or later | Apple silicon and Intel builds are both published |
| Linux | glibc 2.28+, glibcxx 3.4.25+ | Rules out several long-term-support distributions from 2018 and earlier |
| Account | A Google account | Sign-in is needed for model access, not for downloading |
Checked 24 August 2026. Requirements change between preview builds — the official download page is authoritative.