What is Google Antigravity?
Google Antigravity is Google's agent-first coding platform: you give agents a task, they plan, edit code, run commands and test in a browser, and you review the plans and diffs they produce. Google's overview describes it as a set of agentic coding surfaces built on a shared harness co-trained with Gemini models.
The home page puts it this way: "Google Antigravity is our agentic development platform, allowing anyone to build in the agent-first era."
Antigravity is not a single app. As of October 2026 the docs list these surfaces:
| Surface | What it is |
|---|---|
| Antigravity App | A desktop command center for launching, running and inspecting agents, with visual diffs and plan review |
| Antigravity IDE | A standalone editor with the Agent Manager built in |
| IDE extensions | Agent panel, inline completions and diff review inside VS Code, Visual Studio, JetBrains IDEs, Zed and Xcode |
| Antigravity CLI | The agy terminal agent, covered in the Antigravity CLI guide |
| Antigravity SDK | A Python SDK for custom agents, hooks and CI pipelines |
It matters more in 2026 because Google moved a large group of users onto it. The Gemini CLI README says free individual access and Google AI Pro and Ultra (Google One) moved to Antigravity CLI on 18 June 2026, and Google's deprecation page says the Gemini Code Assist IDE extensions stopped serving those users on the same date. Organizations with Gemini Code Assist Standard or Enterprise licenses keep Gemini CLI.
How do you install Google Antigravity?
The getting started page gives the same system requirements for the app and the standalone IDE:
- macOS 12 (Monterey) or later, Apple Silicon or Intel
- Windows 10 (64-bit) or Windows 11, x64 or ARM64
- Linux with glibc 2.28 or later and glibcxx 3.4.25 or later (Ubuntu 20+, Debian 10+, Fedora 36+, RHEL 8+)
- Googlebook, where the app comes preinstalled
For the desktop app, download the installer for your platform: a .dmg to drag into Applications on macOS, an .exe setup wizard on Windows, or a .tar.gz archive on Linux. On Googlebook, install or update through Google Play.
The standalone Antigravity IDE downloads from antigravity.google/download. It updates in the background and shows a "Restart to Update" prompt. Inside the IDE, Cmd+E (Ctrl+E on Windows and Linux) switches between the editor and the Agent Manager.
Sign in with a personal Google account or with Gemini Enterprise credentials when the Antigravity panel asks.
How do you install the IDE extensions?
If you would rather keep your current editor, the IDE getting started tab lists:
| Editor | Where to get it | Requirement |
|---|---|---|
| VS Code | Extensions view, search "Google Antigravity" | VS Code 1.90 or later |
| JetBrains IDEs | JetBrains AI > Settings > Agents, search "Antigravity" | IntelliJ IDEA, PyCharm, WebStorm, GoLand, CLion or Rider 2026.2.1 or later |
| Visual Studio | Extensions > Manage Extensions, search "Google Antigravity" | Visual Studio 2026 on Windows 10 or 11 |
| Zed | agent: open settings, then External Agents > Add > Install from Registry |
Zed 0.140.0 or later, macOS or Linux |
| Xcode | Xcode > Settings > Intelligence, click Get next to Antigravity | Xcode 27 beta 6 or later |
What do the Antigravity plans include?
Google's pricing page lists four ways in, as of October 2026:
| Plan | Price on the page | What it adds |
|---|---|---|
| For Individuals | $0/month | "Access to a variety of agent models", "Unlimited Tab completions", "Unlimited Command requests", "Basic weekly rate limits" |
| Google AI Pro | Not listed on the page | Higher rate limits and a flexible AI credit pool |
| Google AI Ultra | Not listed on the page | Even higher rate limits and a flexible AI credit pool |
| Business plan via Google Cloud | "Standard/Plus plans starting at $30 USD per seat per month" for eligible Gemini Enterprise customers, or a pay-as-you-go plan with a $0 seat fee | Google Cloud terms, spend caps and central admin controls |
The plans page explains the quotas. Users without Pro or Ultra get "a meaningful quota, refreshed weekly." Pro gets a high quota refreshed every five hours until a weekly limit is reached. Ultra gets the highest quota, refreshed every five hours, plus access to third-party models.
Every plan includes "Access to all product features, such as scheduled tasks and the CLI." Pro and Ultra users can buy AI credits for use beyond the quota, and an AI Credit Overages setting (Never or Always) decides whether credits are spent automatically. Google says rate limits depend on capacity and on how much work the agent does, and that bring-your-own-key for extra rate limits is not supported.
Which models does Antigravity use?
The models page lists these as of October 2026:
| Model | Who gets it | Note |
|---|---|---|
| Gemini 4 Argon | Pro (paid or eligible trial), Ultra, Enterprise | Not in the standalone Antigravity IDE; use the app, CLI or extensions |
| Gemini 3.8 Flash | All plans | Tagged Fast in the model selector |
| Gemini 3.7 Flash, 3.6 Flash, 3.1 Pro | Enterprise | Enterprise only |
| Claude Sonnet 5.5 and Opus 5.5 (Thinking) | Pro (non-trial), Ultra, Enterprise | Third-party models |
| Claude Sonnet 4.6, Opus 4.6 and GPT-OSS-120b | Varies by plan | Listed for removal on 2 November 2026 |
You pick a thinking level of Low, Medium or High where a model supports it. The page says your model choice persists between messages in a conversation.
What can Antigravity agents do?
The features page and related docs describe the main pieces of the desktop app:
- Projects. A project can span several folders. When you start a task you choose Local mode, which works in your folders directly, or New worktree mode, which works in an isolated Git worktree so background agents don't collide.
- Artifacts. Agents report through artifacts: implementation plans, code diffs, architecture diagrams, images and browser recordings. You can leave inline comments on a plan to redirect the agent before it changes local files.
- Browser. The
/browsercommand starts a browser subagent. It integrates with Chrome DevTools MCP and records sessions as WebM videos. - Scheduled tasks. Recurring messages to agents on a timer, run with Gemini 3.8 Flash.
- Permissions. By default, "your agent can only read and write within the provided folders of a project," and it asks before running terminal commands outside the sandbox. Broader presets such as Full machine exist when you need them.
- Hooks, terminal and version control. JSON hooks run your own shell scripts at points in the agent loop, and the app has an integrated terminal and a panel to stage, commit and push.
- Remote Control. Drive and monitor desktop sessions from a web browser, with push notifications when a task finishes or needs input.
How do rules work in Antigravity?
Rules are the always-on instructions. The rules page lists these locations for the app, IDE and extensions:
- Workspace:
AGENTS.mdorGEMINI.mdin any folder from the file being edited up to the workspace root, plus.agents/rules/*.md - Global:
~/.gemini/AGENTS.md,~/.gemini/GEMINI.mdor~/.gemini/config/rules/*.md
AGENTS.md and GEMINI.md need no frontmatter and are always active in their folder. Files in .agents/rules/ need YAML frontmatter with a trigger:
| Trigger | When it loads |
|---|---|
always_on |
Full content on every turn |
model_decision |
Only the description loads; the agent reads the rule when the task matches |
glob |
When the agent works on files matching globs |
manual |
When you @-mention it in chat |
Get the frontmatter wrong and, in Google's words, "Antigravity silently discards the rule." A single rule file is cut off after 24,000 bytes, and active global and always-on rules share a 20,000-token budget. To create rules in the app, open Customizations, pick the Rules tab and click + Global or + Workspace. If you already keep an AGENTS.md or GEMINI.md, Antigravity reads it as is.
How do skills and MCP servers work in Antigravity?
Antigravity uses the open Agent Skills format. A skill is a folder with a SKILL.md file, where description is required and name is optional and defaults to the folder name. The skills page lists:
- Workspace (app, IDE, CLI):
<workspace-root>/.agents/skills/<skill-folder>/ - Global (app and IDE):
~/.gemini/config/skills/<skill-folder>/ - Global (CLI):
~/.gemini/antigravity-cli/skills/<skill-folder>/
The agent sees each skill's name and description when a conversation starts and reads the full file only when the task matches. In the app you can also type /<skill-name> to run one by hand. Optional scripts/, examples/ and resources/ folders sit next to SKILL.md.
For MCP, the app has an MCP Store under Settings > Customizations > Installed MCP Servers. The IDE uses ~/.gemini/config/mcp_config.json globally and .agents/mcp_config.json per workspace. Per the MCP page, each server uses command for a local server or serverUrl for a remote one, and remote servers must use Streamable HTTP.
Because .agents/skills/ is a shared convention, the same skill folders work in tools such as Codex and Gemini CLI. If you keep skills and prompts in more than one place, Kipt stores them as Markdown in a private GitHub repo you own, so you can copy a skill folder into .agents/skills/ when you need it. The free SKILL.md generator turns a prompt into a valid SKILL.md in your browser.
How does Antigravity compare with other coding agents?
Antigravity sits in the same space as Cursor and Claude Code: an editor and an agent manager, with a terminal agent alongside. For side-by-side detail, see Claude Code vs Antigravity and Cursor vs Antigravity. If you came from Gemini Code Assist, the Gemini Code Assist guide explains which editions still work and when they end.