How to Use Google Antigravity for Free: The Agent-First IDE Explained
Google Antigravity's multi-agent IDE is free during public preview. Here's how to set it up, what you get for free, and the file-deletion incident worth knowing about first.
Contents4
Antigravity is Google's agent-first IDE, released alongside Gemini 3 and built as a VS Code fork. The pitch is different from Copilot-style autocomplete: instead of one cursor suggesting the next line, you get a Manager view that can run up to 5 agents in parallel across your codebase, each reporting back with screenshots and walkthroughs of what it changed.
It's free right now, in public preview, for individuals. A paid tier is coming eventually, but there's no signup wall stopping you from trying the whole thing today.
Getting it running for free
- Download Antigravity for macOS, Windows, or Linux from Google's official site.
- Sign in with a Google account. No separate API key needed to start.
- Pick your model. Antigravity ships with native support for Gemini 3 Pro, Gemini 3 Flash, and GPT-OSS-120B (OpenAI's open-weights model), plus Claude Sonnet 4.6 and Opus 4.6.
- Switch to Manager view to hand off multi-step work (refactors, test generation, UI iteration) to an agent instead of driving it line by line from Editor view.
- Review each agent's Artifacts (screenshots, diffs, summaries) before merging its changes.
That's the entire cost of entry: a Google account and a download. No credit card, no trial countdown.
What you get for free vs. what's coming
| Available now (free preview) | Expected later | |
|---|---|---|
| Editor + Manager views | Yes | Yes |
| Up to 5 parallel agents | Yes | Likely capped or metered |
| Gemini 3 Pro / Flash access | Yes, included | Possibly usage-based |
| Claude & GPT-OSS model access | Yes | Unclear |
| Paid tier | Not required | Coming, pricing not yet announced |
Google hasn't published what the eventual paid tier restricts, so treat the current free access as a limited-time window rather than the permanent deal.
The catch you should know before turning on autonomous mode
Antigravity's agents can execute file operations on your machine, and that's not a hypothetical risk. One widely reported incident had Antigravity's "Turbo mode" misinterpret a cache-clearing instruction and wipe a developer's entire D: drive, bypassing the Recycle Bin entirely, as covered by both TechRadar and The Register. That's the sharpest edge of giving an agent broad filesystem permissions without a confirmation step.
Beyond that specific incident, community reviews on sites like AlternativeTo describe the current build as buggy and missing features, with some developers calling it "another VS Code fork" chasing what Cursor already built. It's also worth knowing Antigravity reportedly builds on Windsurf's proprietary agent architecture, which Google licensed for close to $2 billion. Some of what feels new under the hood is a licensed, rebranded system rather than something built from scratch.
Should you use it
If you want to try multi-agent coding workflows without paying for [Cursor](https://questloops.com/tools/cursor)'s Pro tier or Claude Code credits, Antigravity's free preview is a legitimate way in, especially if your models of choice (Gemini, Claude) are already supported natively. Just keep backups current and review agent actions before trusting Turbo mode with anything that touches real files you can't afford to lose. Preview software cutting corners on safety checks is a pattern worth remembering, not a one-off.