Google Antigravity
by Google
Google's agent-first development platform: orchestrate parallel coding agents from a desktop app, IDE, CLI or SDK.
Google Antigravity is Google's agentic development platform, where developers delegate coding work to AI agents that plan, edit, run terminal commands and test in a browser, then report back through reviewable artifacts. It targets engineering teams that want multi-agent, asynchronous coding with Gemini, Claude and open-weight models at no entry cost.
Google Antigravity is Google's agent-first software development platform, launched as a free public preview in November 2025 alongside Gemini 3. It grew out of Google's July 2025 deal to hire the Windsurf team, including CEO Varun Mohan, and license its technology for $2.4 billion, and the original product was a VS Code fork pairing an Editor View with tab completions and inline commands with a Manager surface for spawning and observing multiple agents working asynchronously across the editor, terminal and browser. Agents report through Artifacts, meaning task lists, implementation plans, screenshots and browser recordings that developers can comment on without stopping execution. At Google I/O on May 19, 2026, Google shipped Antigravity 2.0, a standalone desktop app for orchestrating parallel local agents with projects, scheduled tasks and subagents, alongside the Antigravity CLI, an Antigravity SDK in preview, Managed Agents in the Gemini API, and enterprise access through the Gemini Enterprise Agent Platform; the classic IDE remains a separate product. Agents work either in the active folder or in an isolated Git worktree. The Individual plan costs $0 and includes Gemini 3.8, 3.7 and 3.6 Flash, Gemini 3.1 Pro, Claude Sonnet and Opus 4.6 and gpt-oss-120b under basic weekly rate limits; Google AI Pro at $20 a month and Ultra at $100 or $200 a month (5x or 20x Pro's tokens) raise those limits, and paid plans unlock /boost, a multi-agent deep-reasoning pipeline. It runs on macOS 12 or later (Apple Silicon only), 64-bit Windows 10 and mainstream Linux. Independent reporting documents real risks: a prompt-injection attack that exfiltrated .env credentials under default settings, an agent that wiped a user's entire D: drive, and recurring quota-exhaustion complaints on Hacker News.
Engineering leaders already standardised on Google Cloud or Gemini who want to pilot multi-agent, asynchronous coding at near-zero entry cost before committing seats elsewhere.
Developers hand off multi-step coding, testing and refactoring tasks to parallel agents and review plans, screenshots and recordings instead of writing every change by hand.
At a Glance
- Category
- Developer Tools
- Pricing
- Freemium, Subscription, Usage-based
- Target Market
- CTOs, VPs of Engineering, Enterprise Developers, Platform Engineering Teams
- Deployment
- Cloud-first, API-based
- Founded
- 1998
- Headquarters
- Mountain View, United States
- Team Size
- 500+
Key Features
- ✓Antigravity 2.0 agent manager
A standalone desktop app to run multiple local agents in parallel, group conversations into Projects, work across workspaces and schedule recurring tasks, so developers supervise rather than type.
- ✓Artifacts for verification
Agents produce task lists, implementation plans, screenshots and browser recordings that users can comment on mid-run, making agent work reviewable instead of buried in raw logs.
- ✓Browser subagent
Agents can drive Chrome to test and verify the applications they build, closing the loop between writing code and checking that the UI actually works.
- ✓Worktree isolation
A New Worktree Mode runs an agent in an isolated Git worktree rather than the active folder, so parallel agents do not trample each other's changes.
- ✓Antigravity CLI
A terminal interface to invoke, monitor and interact with Antigravity agents, run shell commands and manage background subagents without opening the desktop app.
- ✓Antigravity SDK and Managed Agents
An SDK in preview for building custom agents on Antigravity's harness, plus Managed Agents in the Gemini API for programmatic access to the same agent.
- ✓/boost deep reasoning
An on-demand multi-agent pipeline on paid plans that formulates strategy, executes in parallel with verification, then synthesises, aimed at concurrency bugs and multi-file refactors.
- ✓Multi-model choice
The free tier includes Gemini Flash and Pro models alongside Claude Sonnet and Opus 4.6 and gpt-oss-120b, so teams are not locked to a single model family.
Capabilities
Use Cases
- •Parallel feature development
An engineer assigns separate features to several agents in isolated worktrees, then reviews each agent's plan and diff before merging the ones that pass.
- •Hard refactors and concurrency bugs
A developer on a paid plan invokes /boost for an intricate multi-file refactor or race condition, getting a verified multi-agent solution instead of one-shot guesses.
- •Web app build and browser verification
An agent builds a front-end change, drives Chrome to exercise it, and returns screenshots and a recording so the reviewer can confirm behaviour without running it locally.
- •Scheduled background maintenance
A team schedules recurring agent tasks such as dependency checks or routine cleanups that run in the background and surface results for morning review.
- •AI Studio prototype to local codebase
A prototype built in Google AI Studio is exported with its agent code and conversation context into Antigravity to continue as a real local project.
- •Custom internal agents
A platform team uses the Antigravity SDK or Gemini API Managed Agents to embed Antigravity's coding agent into internal tooling and CI-adjacent workflows.
Ideal For
Best For
- ✓Teams on Google Cloud who can buy agentic coding through the Gemini Enterprise Agent Platform under existing Google Cloud terms
- ✓Developers who want to run several coding agents in parallel, each in an isolated Git worktree, and review their output asynchronously
- ✓Front-end and web-app work where a browser subagent can click through and screenshot the result as verification
- ✓Individuals and small teams evaluating agentic coding who need a capable free tier with Gemini, Claude and gpt-oss models
- ✓Prototypes started in Google AI Studio that need to be exported and continued in a local agentic environment
Not Ideal For
- ✗Security-sensitive codebases without strict policy hardening: independent researchers showed default settings (auto-executing terminal, agent-decided review, a browser allowlist including webhook.site) let a prompt-injected agent exfiltrate credentials
- ✗Teams that need predictable, stable tooling: a May 2026 automatic update replaced the IDE interface for some users and wiped local chat history, and HN commenters repeatedly cite Google's record of discontinuing products
- ✗Regulated buyers who need documented SOC 2, ISO 27001, HIPAA or data-residency commitments specific to Antigravity, which Google's Antigravity enterprise page does not list
- ✗Heavy daily users on consumer plans who cannot tolerate hitting rolling 5-hour and weekly quota limits mid-task
Integrations
Deployment
Market Analysis
Pros
- ✓Genuinely usable $0 tier with unlimited tab completions and access to Gemini, Claude and gpt-oss agent models
- ✓Multi-model support was welcomed by HN practitioners who did not want to be locked to Gemini
- ✓Artifacts and browser recordings make agent output reviewable, addressing the trust problem with autonomous coding agents
- ✓Broad surface area in one platform: desktop agent manager, IDE, CLI, SDK and API access, on macOS, Windows and Linux
- ✓Enterprise path under Google Cloud terms, with Google stating corporate data is never used for training when signed in with Google Cloud credentials
Cons
- ✗Security defaults drew criticism: PromptArmor showed a prompt-injected agent exfiltrating AWS credentials from .env files via the browser subagent, and says Google responded with an onboarding disclaimer rather than fixing root causes
- ✗Autonomy can be destructive: How-To Geek reported an agent ran rmdir /s /q on a user's entire D: drive when asked to clear a project cache, with the data largely unrecoverable
- ✗Quota exhaustion is a recurring complaint, from free-tier 'quota limit exceeded' errors at launch to a paid user on HN hitting the 5-hour limit in under 40 minutes and a weekly limit in 4 days
- ✗A May 2026 forced update replaced the IDE workflow with a conversational interface for some users and lost their local chat history
- ✗Reliability doubts: VentureBeat noted early users reported errors and slow generation, and HN users say the agent sometimes claims tasks are complete when they are not
- ✗Platform risk: many HN commenters cite Google's history of discontinuing products as a reason not to build workflows on it
Pricing
Individual
$0
- ✓Agent model access to Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.1 Pro, Claude Sonnet & Opus 4.6, gpt-oss-120b
- ✓Unlimited Tab completions
- ✓Unlimited Command requests
- ✓Basic weekly rate limits
Google AI Pro
From $20/mo
- ✓Everything in Individual
- ✓More generous rate limits
- ✓Flexible AI credit pool
- ✓/boost deep reasoning (paid plans)
Google AI Ultra
From $100/mo
- ✓$100/mo tier with 5x the tokens of Pro
- ✓$200/mo tier with 20x the tokens of Pro
- ✓More generous rate limits
- ✓Flexible AI credit pool
Organization (via Google Cloud)
Contact for pricing
- ✓Access under Google Cloud Terms of Service
- ✓Google Cloud Project integration
- ✓Consumption-based API pricing with Gemini Enterprise Agent Platform
- ✓Included in select Gemini Enterprise app subscriptions
Antigravity itself is free; capacity is bought through Google AI subscriptions. On May 19, 2026 Google kept Pro at $20/month, added a $100/month Ultra tier with 5x Pro's tokens and cut top Ultra from $250 to $200/month for 20x, pooled Flash and Pro quota, and made AI credits overage protection rather than a base entitlement. Organizations pay consumption-based Gemini Enterprise Agent Platform API rates under Google Cloud terms; no per-seat list price is published.
Security & Compliance
Sources
This page was written from 16 sources, 9 on domains other than antigravity.google.
- 1.antigravity.google — antigravity.googlevendor
- 2.antigravity.google — pricingvendor
- 3.antigravity.google — changes to antigravity plansvendor
- 4.antigravity.google — google io 2026vendor
- 5.antigravity.google — enterprisevendor
- 6.antigravity.google — getting startedvendor
- 7.antigravity.google — boostvendor
- 8.developers.googleblog.com — build with google antigravity our new agentic development pl
- 9.venturebeat.com — google antigravity introduces agent first architecture for a
- 10.techcrunch.com — google launches antigravity 2 0 with an updated desktop app
- 11.promptarmor.com — google antigravity exfiltrates data
- 12.howtogeek.com — google antigravity ide deleted someones entire drive
- 13.0xsid.com — antigravity bait n switch
- 14.hn.algolia.com — 45967814
- 15.hn.algolia.com — 48222529
- 16.hn.algolia.com — 49517537
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Raindrop
AI agent monitoring that catches silent production failures: Sentry for AI agents
GitLab Duo Agent Platform
Agentic AI across the whole GitLab DevSecOps lifecycle: planning, coding, code review, CI/CD and security agents under one governance model
CodeRabbit
AI code review and agentic change management for teams shipping human- and machine-written code
Qodo
Agentic code review and governance layer for teams shipping AI-generated code at scale