Google Antigravity 2: Agent-First Dev Platform with Gemini 4 Argon

Google Antigravity is Google’s agent-first development platform for building software with autonomous AI coding agents. It can investigate a codebase, implement features, debug problems, run development commands, and inspect a website in Chrome. Multiple agents can work on different tasks across projects, while the desktop app keeps their conversations, progress, and results in one place.

Antigravity 2.0 expands the original AI coding editor into a family of developer products. The standalone desktop app manages ongoing agent work, Antigravity CLI brings those capabilities to the terminal, and IDE integrations keep coding assistance close to an editor. Google has also introduced a Python SDK for custom agents and Gemini 4 Argon for demanding engineering tasks on eligible plans.

Features

• Run several coding agents across projects, repositories, and independent Git worktrees.
• Delegate research, implementation, and debugging work to parallel subagents.
• Let agents read and edit files, execute terminal commands, and search the web.
• Inspect code changes, implementation plans, screenshots, and browser recordings as artifacts.
• Use Chrome to test website interactions and inspect UI changes.
• Schedule recurring agent tasks and continue conversations remotely.
• Work through the desktop app, terminal CLI, AI-powered IDE, or editor extensions.
• Extend agents with Skills, custom subagents, plugins, hooks, and MCP servers.
• Enable Google integrations for Firebase, Android, Chrome DevTools, Flutter, and cloud development.
• Dictate prompts using live voice transcription in the desktop app.

How Google Antigravity Works

In Antigravity, an agent receives a development goal and uses the project files and connected tools to work through it. For example, a feature request can involve reading components and API handlers, changing code, running a test suite, opening the result in a browser, and preparing a walkthrough. You interact with the agent through a conversation and the work it produces, including file diffs and other artifacts.

Projects and Parallel Agents

Projects group the code, settings, and agent conversations for a task or product. One Project can include several folders or repositories, such as a React frontend and a Node.js API. An agent can reason across those folders while each Project maintains its own configuration and conversation history.

Antigravity lets you launch multiple agents and switch between their conversations while tasks continue. A lead agent can also create focused subagents to inspect a test failure, search a repository, or investigate a subsystem. These tasks can run asynchronously, which is especially useful when a feature touches several parts of a codebase.

For Git repositories, the desktop app offers Local mode and New Worktree mode. Local mode works directly in your existing checkout. New Worktree mode opens an independent Git worktree for an agent conversation. That gives concurrent tasks their own working copies of the repository and keeps their file changes organized.

Artifacts, Feedback, and Browser Testing

Agents communicate their progress through artifacts. An implementation plan describes the intended change, a code diff shows edited files, and a walkthrough records the finished work. Artifacts can also include architecture diagrams, screenshots, and browser recordings. The desktop app provides a dedicated review pane. The CLI has a keyboard-driven artifact review experience.

Antigravity agents can interact with Chrome after editing a web project. They can open pages, test interactions, capture screenshots, and record browser activity. This is useful for tracing UI failures and checking the result of a change in the running app. You can comment on plans and artifacts during an agent conversation to request a different implementation or clarify the task.

Scheduled Tasks and Remote Control

Scheduled Tasks run agent prompts on a recurring schedule, such as checking a repository for outdated dependencies or collecting a daily summary of test failures. Antigravity also supports persistent background processes called sidecars for recurring jobs and event-driven work. Regular maintenance can continue apart from an interactive coding conversation.

Remote Control lets you access supported desktop and CLI sessions through a web browser on another computer, tablet, or phone. The session keeps running on its host computer, and the remote dashboard gives you access to its ongoing agent conversations. This is useful for checking a long-running task when you are away from your development machine.

Antigravity App, CLI, IDE, Extensions, and SDK

Google provides several ways to work with the Antigravity agent system. The desktop app is the main workspace for coordinating independent agent tasks. The CLI suits terminal-based development and remote shells. The IDE and editor extensions serve developers who prefer editing code in their normal development environment. The SDK exposes the agent harness to Python applications.

Antigravity CLI

Antigravity CLI runs the agent harness in a terminal UI. Launch it with agy, describe your task, and work with files and shell commands from your repository. Its keyboard-oriented experience includes conversation history, slash commands, subagents, plugins, and code review. It also supports noninteractive execution for scripts and automated jobs.

The CLI uses the same core agent harness as the desktop app, and you can import compatible terminal conversations into the desktop workspace. It supports slash commands for tasks such as inspecting a diff, managing subagents, resuming a conversation, and checking model usage. Headless mode runs prompts from scripts and automation jobs.

IDE and Editor Extensions

The original Antigravity IDE remains available for developers who prefer a full editor. It includes coding assistance, agent conversations, and codebase-aware editing. Separate Antigravity extensions bring agent capabilities to established editors, including Visual Studio Code, Visual Studio, JetBrains IDEs, Zed, and Xcode. Model availability and enterprise sign-in differ across these products.

Antigravity SDK

The Antigravity SDK is a Python framework for building agents on Google’s Antigravity harness. Developers can configure a model, register Python tools, define agent personas, create subagents, use MCP integrations, and receive structured results. The SDK also supports lifecycle hooks and local-model configurations. It is intended for developers writing their own agent applications, beyond using the ready-made desktop and CLI products.

Gemini 4 Argon and Other AI Models

Gemini 4 Argon is now available in Google Antigravity since October 13, 2026. It is intended for demanding, multi-step engineering work: tracing failures through large codebases, investigating logs and commit history, planning changes across multiple modules, and running follow-up tests. An Argon agent can investigate a bug report, find the affected code, check whether an earlier fix reached production, and prepare a documented resolution.

Within Antigravity, Gemini 4 Argon requires Google AI Ultra or an eligible Gemini Enterprise subscription. It works in the Antigravity desktop app, CLI, and compatible IDE extensions. The original Antigravity IDE does not include Argon. Google AI Pro provides Argon access in certain other Google products, but Google AI Pro does not unlock Argon in Antigravity.

Gemini 3.8 Flash is available for everyday agent tasks on every Antigravity plan. The model picker also includes Anthropic Claude and GPT-OSS options according to your subscription. Model choice persists within a conversation until you change it.

Skills, Plugins, and Google Integrations

Antigravity supports Agent Skills for reusable instructions, custom subagents for specialized work, MCP servers for external tools, and hooks for actions triggered during the agent lifecycle. Plugins package related skills, agents, rules, hooks, and MCP configurations. You can use these features to teach an agent project conventions or connect it to systems it needs during development.

Build with Google integrates platform-specific Skills, MCP servers, and development utilities into Antigravity. An agent can use these connections while working on the project, including operations against Firebase services, Android emulators, browser diagnostics, and Google Cloud development tasks..

What Can You Build with Google Antigravity?

Full-Stack Websites and Applications

An agent can work through a feature request that touches UI components, API endpoints, and tests. For a dashboard, you might ask it to build a settings screen, connect a form to an existing endpoint, run the project’s tests, and check the resulting page in Chrome. Projects that include frontend and backend repositories give the agent context from both parts of the application.

Frontend UI Development and Debugging

Frontend work often requires checking behavior in a running browser. Antigravity can edit a React or other web project, launch the development server, navigate the page, and record what happened. The resulting screenshots and recordings help illustrate layout changes, broken interactions, and the effect of a fix.

Investigating Bugs Across a Codebase

For a failing test or recurring production issue, an agent can inspect the relevant files, follow imports or call sites, examine logs, and run local checks. Parallel agents can examine independent hypotheses or subsystems. This also works for cross-repository investigations when a UI problem may originate in an API or shared library.

Codebase Modernization and Migration

For an older application, an agent can trace dependencies, inspect build or test failures, map a migration, and edit related modules across a repository. Parallel agents can investigate different subsystems before the main implementation proceeds. In one internal Google engineering project, Argon traced a YouTube comments performance problem to repeated emoji rendering. The resulting 12 optimizations cut the comments frontend CPU footprint in half. In another project, an Argon agent migrated legacy scientific code from Fortran to Rust. Google reported a 1.8× performance improvement for that migration.

Recurring Repository Maintenance

Scheduled agent tasks can inspect dependency changes, run routine checks, or prepare summaries of repository health. This suits work that recurs on a weekly or daily cadence. Developers can check the resulting conversations and artifacts when returning to the desktop app or via Remote Control.

Technical Research and Documentation

Antigravity agents can read source files, browse technical references, and prepare architecture notes or implementation proposals. That is useful when joining an unfamiliar project or planning a change across several modules. The artifact system keeps plans and diagrams alongside the conversation that produced them.

How to Get Started

The standalone Antigravity desktop app is available for macOS, Windows, and Linux. Install it from the Google Antigravity website and sign in with a Google account or an eligible enterprise account.

1. Create a Project in the desktop app and associate a local folder or Git repository. A Project can also include several related repositories.
2. Open a conversation, select an available AI model, and choose Local or New Worktree mode for a Git checkout.
3. Describe the desired feature, bug, or investigation. Include the relevant files, expected behavior, or acceptance criteria when they are known.
4. Follow the conversation and open its plans, file changes, or browser artifacts as the agent works.
5. For recurring work, configure a Scheduled Task. For terminal use, install Antigravity CLI and run agy from your project.

For example, a first task in an existing web app could be: “Build a responsive account settings page using the components already in this repository. Connect the form to the existing settings API, run the relevant tests, and check the page in Chrome.”

Pricing and Usage Limits

Google Antigravity has a free individual tier. Free accounts can use the desktop app, CLI, scheduled tasks, Gemini 3.8 Flash, and unlimited Tab completions. The free agent allowance refreshes weekly. A short code change typically consumes less of that allowance than a multi-step investigation.

Google AI Pro has a higher agent allowance that refreshes every five hours until the weekly quota is reached. Google AI Ultra has the highest individual allowance, five-hour refreshes, higher weekly capacity, and Argon access in Antigravity. Gemini Enterprise uses eligible business subscriptions with enterprise-specific model access. Pro and Ultra subscribers can also choose to spend purchased AI credits after exhausting a model’s included allowance. This option is controlled in settings and is not charged automatically when set to Never.

The Antigravity SDK is a separate developer library. Calling paid model APIs from a custom SDK application can incur API charges outside Antigravity’s desktop and CLI subscription allowances.

Recent Updates

• October 13, 2026: Gemini 4 Argon rolled out to Google AI Ultra Antigravity desktop and CLI users and eligible enterprise customers.
• October 12, 2026: Antigravity app v2.23.0 introduced a permissions picker in the prompt box, an expanded media viewer, and ZIP/folder attachments.
• Summer 2026: Remote Control, IDE extensions, and product integrations expanded the ways developers interact with agent sessions.
• May 2026: Antigravity 2.0 introduced a standalone agent-management desktop app and the companion CLI.

Pros

• A free tier for trying autonomous coding agents on real projects.
• Parallel agent conversations and independent worktrees for larger tasks.
• Plans, code diffs, screenshots, and browser recordings in one development experience.
• Desktop, terminal, editor, and Python SDK options within the Antigravity ecosystem.
• Google developer integrations for web, Firebase, Android, and cloud projects.

Cons

• The agent-first desktop workflow takes adjustment for developers accustomed to working directly in an editor.
• Heavy multi-agent work consumes usage quotas, with higher allowances tied to paid plans.
• Gemini 4 Argon access in Antigravity requires Google AI Ultra or an eligible enterprise plan.

FAQs

Can Google Antigravity work on multiple repositories?

Yes. A Project can include multiple folders and Git repositories, giving agents context from related frontend, backend, or shared-code projects. Agents can work in local checkouts or independent Git worktrees.

Does Google Antigravity work offline?

The desktop app and CLI require online access for Google-hosted agent models and account authentication. The Python SDK can be configured for certain local models.

Can I access an Antigravity agent from my phone?

Remote Control provides browser access to supported agent sessions from a phone, tablet, or another computer. The host computer runs the original session and must stay connected.

What happened to Gemini CLI?

Google transitioned free and individual Google AI Pro/Ultra Gemini CLI access to Antigravity CLI in June 2026. Gemini CLI continues for supported enterprise and API-key configurations. Existing Gemini CLI users can migrate compatible settings and customizations to Antigravity CLI.

Is Antigravity an IDE or an agent manager?

The Antigravity product family includes both. Antigravity 2.0 is a standalone desktop agent manager, while Antigravity IDE remains a full coding editor. Antigravity CLI and editor extensions provide additional ways to work with the agents.

Alternatives and Related Resources

• Antigravity CLI: Gemini CLI Alternative for Agent-first Development
• Antigravity CLI Cheatsheet: Commands, Setup, Shortcuts, and Workflows
• 10 Best CLI AI Coding Agents in 2026: Open-Source & Commercial
• Google AI Timeline: Gemini 4 Argon, Models & Product Releases

Official Sources

• Google Antigravity documentation
• Google Antigravity changelog
• Gemini 4 Argon in Google Antigravity

More Google AI Tools

• Code Wiki: Google’s AI-Powered Solution for Code Understanding
• Gemini Notebook (NotebookLM): Free Google AI Research Assistant
• Google Stitch: Free AI Vibe Design Tool with Agent and Web Export
• Pomelli: Google’s Free AI Tool for SMB Marketing Campaigns
• Google Flow Music: Free AI Song Generator Powered by Lyria 3.5