What Is Google Antigravity? Agent-First IDE Explained: Plans, Models, Permissions and Release History

Google Antigravity is Google's agentic development platform: a desktop app, an IDE, a command line agent, IDE extensions and a Python SDK that let AI agents plan, write, run and test code with a developer reviewing the results. It launched in public preview on 18 November 2025 and was rebuilt as Antigravity 2.0 on 19 May 2026, so a lot of what is written about it describes a product that no longer looks the same. This page sets out what Antigravity is today, how its plans and quotas work, which models it offers, how its permission system works, what enterprise customers get through Google Cloud and how it has changed since launch. It is written for developers, engineering managers and the people who approve tools for a team. Every fact comes from Google's own Antigravity website, documentation, pricing page, terms, blog and changelog. Last checked: 2026-11-15.

This is an independent guide from NHI Mgmt Group, based on Google's public Antigravity documentation as of November 2026. It is not sponsored or reviewed by Google. Product features change, so confirm settings against Google's current documentation.

Key takeaways

• Five surfaces, one sign-in. Antigravity is now an App, an IDE, a CLI called agy, extensions for six IDEs and a Python SDK, all sharing one sign-in that decides models, quotas and data terms.
• Free for individuals, quota based. The pricing page lists an individual tier at $0 a month with unlimited Tab completions and a weekly rate limit; Google AI Pro and Ultra add higher limits and a credit pool.
• Business use runs through Google Cloud. Organisations buy through Gemini Enterprise: Standard or Plus seats from $30 per seat per month, or a pay-as-you-go plan with no seat fee.
• Gemini first, with partner models. The models page lists Gemini 4 and Gemini 3.x models plus Claude and GPT-OSS models, with availability that differs by plan and some models due to be removed on 18 November 2026.
• Data terms depend on the account. Individual terms allow Google to use interactions to improve products unless Telemetry is turned off; enterprise documentation states that code, prompts and transcripts are not used to train Google foundation models.

What is Google Antigravity?

Google's documentation describes Antigravity as a set of agentic coding tools that provide "native agentic assistance across purpose-built product surfaces", supporting both turn by turn pair programming and asynchronous background agents. The home page calls it Google's agentic development platform for "the agent-first era".

The product started as an IDE. The launch post on 18 November 2025 introduced two views: an Editor view with Tab completions, inline commands and an agent side panel, and an Agent Manager for spawning and monitoring several agents across workspaces in parallel. Agents produced Artifacts (task lists, implementation plans, walkthroughs, screenshots and browser recordings) that the developer could review and comment on, and could drive a browser to test their own work.

Antigravity 2.0, announced on 19 May 2026, moved the Agent Manager idea into a separate desktop application with no editor. Google's 2.0 post says the App "retains many of the core principles" of the Agent Manager but is "a completely separate desktop application", grouped by Projects rather than repositories. The same day Google released the Antigravity CLI, the SDK and an API. The 2.0 post states that the Agent Manager inside the IDE stays available for now and is planned to be removed in an upcoming release.

Use cases on the site cover frontend and full stack work, legacy migrations, testing, incident triage and code review, and since 9 November 2026 a Knowledge Work layout aimed at product, finance, design and other non-engineering roles.

Key facts at a glance

As of 15 November 2026, from Google's Antigravity pages.

Plans and pricing

The table follows antigravity.google/pricing and the plans documentation as of 15 November 2026. Prices are in US dollars. The Antigravity pages do not show prices for Google AI Pro or Ultra, which are Google consumer subscriptions sold separately.

How limits work: the plans page says baseline limits "depend mainly on available capacity", track the agent's workload (simple tasks allow more prompts than complex ones) and may change. Pro and Ultra users can buy AI credits, consumed at standard Gemini Enterprise consumption pricing, and choose an AI Credit Overages setting of Never or Always to decide whether credits are spent once a model's baseline quota runs out.

What is not offered: the plans page states that bring your own key or endpoint for extra rate limits is not supported, and that there are no organisational tiers through a custom contract outside Gemini Enterprise.

Enterprise billing details: third-party models are billed by token consumption and need administrator opt-in, and first-party background models that coordinate work also consume tokens. Gemini Enterprise seat holders do not need a separate per-user Antigravity licence.

Models available

Antigravity's models page lists 14 selectable models as of 15 November 2026 and states that they come from "Google DeepMind and partner providers". It gives no context window figures, so none are given here.

• Image generation: Nano Banana 2 powers the generative image tool and is not user-selectable.
• Two pages, two wordings: the plans page says third-party models are Ultra only, while the models page marks the Claude 5.5 models as available on non-trial Pro subscriptions. The table follows the more detailed models page; check both before relying on Pro access.
• Enterprise opt-in: partner models must be turned on by an administrator in Gemini Enterprise settings before enterprise users see them.
• Reasoning commands: the overview lists /boost for deeper reasoning and /teamwork for multi-agent work, and changelog v2.31.0 adds /swarm for coordinating subagents.

Features and modes

Supported IDEs and platforms

• Desktop: the App and IDE download for macOS (Apple Silicon and Intel), Windows and Linux; the home page also lists Googlebook. Windows Subsystem for Linux support arrived in App v2.16.0.
• IDE extensions: Visual Studio Code (VS Code Marketplace), Visual Studio 2026 (VSIX), JetBrains IDEs such as IntelliJ IDEA, PyCharm, WebStorm, GoLand, CLion and Rider, Zed, Xcode and Eclipse (Public Preview). Extensions were announced on 20 August 2026.
• Enterprise status of extensions: enterprise support is in Preview for JetBrains, Zed, Xcode and Eclipse.
• Other editors: any editor that speaks the Agent Client Protocol can launch agy acp; the docs list VS Code, Zed, JetBrains IDEs, Android Studio and Xcode, with features depending on the editor.
• Standalone IDE and enterprise: the enterprise documentation states that the standalone Antigravity IDE is not supported for enterprise deployments; enterprises use the App, CLI and IDE extensions.

Enterprise and admin controls

Permission presets in the App: Request Review (no sandbox, always ask before commands, workspace only), Default (the documented default: sandbox on, commands allowed inside the sandbox and asked outside), Auto (sandbox on, risky commands outside it checked) and Turbo (no sandbox, unrestricted commands, full file system). Auto is not offered to business and enterprise accounts, and Request Review is the fallback when the sandbox is unavailable.

Access lists: rules take the form action(target) with actions such as read_file, write_file, read_url, command and mcp, held in Deny, Ask and Allow lists with precedence Deny > Ask > Allow. The CLI keeps them in ~/.gemini/antigravity-cli/settings.json and its toolPermission policy defaults to request-review. Inside the sandbox, terminal commands are limited to the workspace and temporary directories with no network access by default, and read_url grants become the outbound network allowlist.

Application Default Credentials and Workforce Identity Federation configurations are non-human credentials once they drive headless agents: give each an owner and scope. See AI Coding Tools and Secrets Sprawl and How to audit AI coding assistant tokens and config files.

Data handling and privacy

• Individual accounts: the Antigravity terms (Section 5) state that Google uses "Interactions" (user data, interaction data, metadata and feedback) to evaluate, develop and improve Google and Alphabet research, products, services and machine learning technologies. Users can change this in settings through the Telemetry toggle, but Google still uses data to provide the service and keep it safe with Telemetry off.
• Human review: the terms state that Google employees and contractors may access and review Interactions, and advise against submitting confidential information.
• Deletion: no fixed retention period is stated; users can request deletion of Interactions by email, and delete Remote Agents data in settings.
• Enterprise accounts: the IDE extensions page states that for Gemini Enterprise sign-ins, code, prompts and transcripts are never used to train Google foundation models, and that IAM policies, VPC Service Controls and dual-region or sovereign cloud boundaries are respected. Use falls under the Google Cloud Terms of Service.
• Residency: regional data residency applies to first-party Gemini models; third-party model inference runs in global, us and eu locations, and in-country regions are not supported.
• Third-party models: the terms (Section 8) state that a third-party or open-source model's own terms apply when you choose it.

The practical consequence: a developer signed in with a personal Google account on a work repository is under consumer terms, not the enterprise terms. Decide which sign-in is allowed before rollout. See Unsanctioned AI Coding Assistants.

Release history

Dated announcements from Google's Antigravity blog and the App changelog. The changelog page shown covers App releases from v2.14.0; earlier releases are not summarised here.

Limits and known constraints

• Capacity-based limits: rate limits depend on available capacity and agent workload and may change, so the same plan can allow a different number of prompts from week to week.
• No extra limits through your own key: bring your own key or endpoint is not supported for additional rate limits.
• Model removals: Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS 120B are marked for removal from the plans that carry them on 18 November 2026.
• Enterprise surfaces: the standalone IDE is not supported for enterprise deployments, and enterprise support for several IDE extensions is in Preview.
• Business account presets: the Auto preset and CLI auto mode are unavailable to business and enterprise accounts; unsandboxed rules are available only to them.
• Application Default Credentials: models older than Gemini 3 Flash are unsupported, and image generation is only on global endpoints.
• Agent Manager in the IDE: Google has said it plans to remove it in an upcoming release.

Security guidance

This page describes what Antigravity is and costs. For hardening, apply the discipline in the AI IDE and Coding Agent Security Guide. The Antigravity-specific points: keep the Default or Request Review preset for work repositories and treat Turbo, which turns off the sandbox and allows unrestricted commands, as an isolated-environment setting only, as the enterprise docs advise for unrestricted auto-approval. Write Deny rules for credential paths such as SSH keys and .git/, keep read_url grants narrow because they open the sandbox's network, and review MCP servers and plugins before installing them. Treat repository rules, skills and hook files as untrusted input; see Coding Agent Configuration File Attacks and the MCP Security Guide.

Practitioner checklist

• Decide whether staff use personal Google accounts or Gemini Enterprise sign-in, and block the consumer sign-in on work code if you need enterprise data terms.
• Record the date and plan whenever you quote Antigravity limits or prices, and recheck the pricing and plans pages before renewal.
• Enable AI developer tools only in the regions you need, and opt in to partner models deliberately after reviewing each model provider's terms.
• Set a global permission preset of Default or Request Review, and review per-project overrides that move away from it.
• Maintain Deny lists for secrets, SSH keys and .git/ writes, and keep read_url allowlists to named domains.
• Put Antigravity's Google Cloud services inside VPC Service Controls and route request and response logs to your SIEM.
• Inventory Application Default Credentials and Workforce Identity Federation configurations used for headless agents, with an owner for each.
• Plan for model removals such as those dated 18 November 2026 before pinning model names in internal guidance.

Standards and references

• Google Antigravity: Documentation overview
• Google Antigravity: Pricing
• Google Antigravity: Plans
• Google Antigravity: Models
• Google Antigravity: Permissions
• Google Antigravity: Enterprise
• Google Antigravity: Terms of Service
• Google Antigravity: Changelog

Related NHI Mgmt Group resources: AI IDE and Coding Agent Security Guide · AI Coding Tools and Secrets Sprawl · Coding Agent Configuration File Attacks · AI Coding Assistant Buyer's Guide