If you're looking for a modern IDE that combines code editing, automation, and AI-powered verification, Antigravity es Just what you were waiting for. This Google environment was created with an agent-first approach.This means it doesn't just suggest lines: it orchestrates agents capable of planning, executing, and validating complex tasks without you having to be on top of it all the time.
In this detailed guide You will learn how to install Antigravity on Linuxto fine-tune it and take advantage of it with real-world use cases. You'll see how its Agent Manager works, how it integrates with Chrome, and what artifacts it generates. to give confidence in every change and why it makes a difference compared to traditional chat-based IDEs.
What is Google Antigravity and why does it matter?
Antigravity is a agent-based development platform that redefines the IDE in terms of autonomy and verification. Its main interface is the Agent Manager (Mission Control)From which you launch, monitor and coordinate multiple agents working in parallel on different tasks and workspaces.
Instead of relying on a linear chatbot-type interaction, Antigravity allows you to create multiple agents that attack multiple problems at once.Generating and validating results asynchronously. Code editing is still available and maintains the familiarity of VS Code, but the priority shifts to engineering management and output quality.
The proposal is based on models such as the Gemini 3 Pro and, according to the documentation itself, It arrived as a Public Preview geared towards personal accounts. In Google's words, it's a leap from "Copilot" to "Coworker": a digital coworker that plans, schedules, and checks itself through the browser and local tools.
Requirements and availability
To begin with, Antigravity is available for Linux, Windows, and macOS. You will need a personal Gmail account and the Google Chrome browser (or Chromium-based version)., given that some of its capabilities depend on close integration with the browser.
The preview can be used for free with limited fees, including access to Gemini 3 Pro and, in the individual plan mentioned in the consulted documentation, to Claude Sonnet 4.5 y GPT-OSSPlease note that when you reach the quota limit, the IDE itself will display warnings; It is advisable to choose the model with the head to balance quality and consumption.
Installing Antigravity on Linux
In Debian/Ubuntu-based distributions, Antigravity maintains its own repository. Repo registration and installation are straightforward. with APT:
sudo mkdir -p /etc/apt/keyrings curl -fsSL https://us-central1-apt.pkg.dev/doc/repo-signing-key.gpg | \ sudo gpg --dearmor -o /etc/apt/keyrings/antigravity-repo-key.gpg echo "deb [signed-by=/etc/apt/keyrings/antigravity-repo-key.gpg] https://us-central1-apt.pkg.dev/projects/antigravity-auto-updater-dev/ antigravity-debian main" | \ sudo tee /etc/apt/sources.list.d/antigravity.list > /dev/null sudo apt update sudo apt install antigravity
For distributions with RPM packages (Fedora, RHEL, etc.), You can use DNF with this repo:
sudo tee /etc/yum.repos.d/antigravity.repo << EOL [antigravity-rpm] name=Antigravity RPM Repository baseurl=https://us-central1-yum.pkg.dev/projects/antigravity-auto-updater-dev/antigravity-rpm enabled=1 gpgcheck=0 EOL sudo dnf makecache sudo dnf install antigravity
Users with an Arch-based distribution can download it from the AUR. Being new software, it's not easy to say which is the best option, but the package antigravity-bin It is the one with the best rating.
Once installed, run Antigravity. The initial wizard will guide you through the setup.You can import settings from VS Code/Cursor or start from scratch, choose a theme (dark or light), and define how you want to interact with the agent.
Initial start and autonomy policies
During onboarding you will see two key parameters: the Terminal execution policy and Review policyAdjusting them properly prevents surprises, especially in sensitive environments.
- Terminal execution:
- off: never executes commands automatically, except for a permitted list.
- CarsThe agent decides and asks for permission when he deems it necessary.
- Turbo: executes commands automatically (with configurable blocklist).
- Demo reel:
- Always Proceed: never requests review.
- Agent DecidesThe agent requests a review when he deems it appropriate.
- Request Review: always requires review before proceeding.
Based on these policies, there are four predefined usage flows: Agent-driven, agent-assisted, review-based development y Custom settingsThe recommended option is Agent-Assisted Development because it balances autonomy and control, asking for confirmations when necessary.
Complete the rest of the wizard: Choose your Editor preferencesSign in with your Google account and review the Terms of Use. During the login step, Antigravity opens Chrome and creates a dedicated profile; This allows its creation so that the managed browser functions correctly..
Agent Manager: the control center
When you open Antigravity you will see the Agent Managernot the file tree. This view is designed to orchestrate asynchronous agents that work on different tasks, with a clear visualization of the status, generated artifacts, and pending reviews.
You can open a Workspace (local folder) from here or later. Select a working folder and Antigravity will initiate a contextual conversation about that space. Additionally, you can use @ to reference files and provide extra context to your statement.
In the top bar you will see two important controls: Planning vs Fast, and the model selectionPlanning produces comprehensive plans and artifacts (ideal for complex tasks), while Fast executes directly (perfect for small changes when speed is paramount). Keep in mind the Gemini 3 Pro installmentsIf you run out, the IDE will tell you.
What's in the Agent Manager interface
The view is organized into functional sections. Knowing them greatly speeds up navigation between tasks and conversations.
- InboxLists of active conversations. You can enter at any time to view messages, status, artifacts, and approval requests.
- Start conversation: Start a new conversation directly, with the Ask anything box ready.
- Workspaces: Add and select workspaces where the agent can create and modify files.
- Playground: area for rapid experimentation; later you can turn it into a Workspace if the work matures.
- Editor View: Switches to Editor view to view and edit files, launch online commands, and interact with the agent alongside the code.
- Web Browser: access to the integrated browser, one of Antigravity's major differentiators by allowing "end-to-end" verification.
When you're comfortable, click Next in the introductory dialogs and Open your first Workspace or stay in Playground for quick tests. If you come from other AI environments, the mention of AI will be familiar. @archive and the context pass.
Configure the Antigravity browser
The main agent can invoke a browser sub-agent which operates open pages in a Chrome managed by Antigravity (a specialized model distinct from that of the main agent). This subagent will click, write, read the DOM and the consoleas well as capturing screens and recording video.
To enable it, Antigravity guides you through the installation of its Chrome extension. For example, in Playground, write a command like Go to antigravity.google; The agent will request to configure the browser. and it will display a Settings button to install the extension from the Chrome Web Store.
With the extension installed, the agent will ask for permission to continue the task. You'll see the browser "come to life" with a blue border indicating agent control and, upon returning to the Agent Manager, states and artifacts related to that navigation will appear.
Artifacts: closing the trust gap
A key part of the agent-first approach is its artifacts. Instead of assuming that he "fixed the bug," the agent demonstrates with evidence What he did and how he validated it. You can activate the Artifacts view from the top corner.
- Task Lists & PlansThe agent plans the work in clear steps that are editable and can be approved by you.
- Code Diffs: standardized differences of which lines change and where.
- Screenshots: screenshots of the UI before and after.
- Browser Recordings: full videos of the agent's actions in the browser.
- Test Results: structured records of tests performed with their results.
Also, you can Comment on artifacts in the style of Google Docs and send granular instructions: the agent will incorporate them and iterate again. This is where the role of “architect” comes into play, providing precise review and guidance.
The Editor: context, online commands, and agent panel
The Editor retains the VS Code experience: file explorer, syntax, extensions And, in general, the muscle memory you already have if you come from that background. However, it's enhanced with awareness of the agent.
- Inline Command: select code and issue commands such as "make it more efficient" or "add comments to this logic".
- Agent Side PanelOn the right side, open conversations and request changes to the code with immediate context.
To switch between Editor and Agent Manager, you have buttons in the upper right corner (Open Editor / Open Agent Manager) Y, If you prefer the keyboard, the shortcut is Cmd + E (or its equivalent in your system).
Guided use cases that demonstrate its power
Featured news with navigation and summary
A simple example: ask the agent to visit Google News and summarize relevant headlines. From Agent Manager, enter Playground and makes a request like: visit a certain site and bring back a summary by categories.
The agent will launch the browser, take control, and generate artifacts such as screenshots and video. Also try a more specific instruction: Visit https://docs.cloud.google.com/release-notes and get me a summary of the release notes, categorized by product.
Generate a dynamic website in Python and Flask
Imagine an informational website for a one-day technical conference. Request it from the agent with complete instructions (home page with date and location, 8 talks, speakers, searches by category/speaker/title, 60-minute break, dummy data, and detailed README). Use the Python + Flask stack, HTML/CSS/JSand requests that the website be launched for review.
The agent will produce several artifacts: Task Plan, Implementation Plan, and Walkthroughwith browser verification included. If a port is busy (e.g., 5000), the agent will look for another free port (such as 8080) and will continue without blocking.
Once the website is live, switch to the Editor to view the generated files. Do you want to add two more talks? From the agent panel, request "Add two more talks to the schedule" and review how to update the plan and validate the changes.
Productivity app: Pomodoro timer
Another classic front-end test: a Pomodoro app with a polished aesthetic. It asks for “Create a productivity app that features a Pomodoro timer. Give a calm and aesthetic look to the application.” and observe the process. The agent will run tests and generate a media artifact with a video. browser verification.
If something visual doesn't fit (for example, an icon with a non-transparent background), suggest changes:Make the hourglass icon background transparent". The agent will iterate the styleIt will update the plan and show you the differences.
Unit testing with simulated stubs
Open a Workspace with a Python module that implements the order flow (class Order) with custom exceptions, external services Inventory Service y Payment Gatewayand logic of checkout that validates cart, stock, discounts, payment and inventory decrease. It's a perfect scenario for generating tests with mocks..
Use @ to reference the file within the conversation and ask: “generate unit tests for this module and test it out with mock implementations". The agent will create artifacts with the test planIt will implement stubs, run the tests, and show you the results. You can also request a visualization of the class to better understand its design.
Return to your inbox to resume tasks
If you've started multiple conversations, go to Inbox to see them all together. From there you can pick up the conversation where you left off., review artifacts and states, or give new instructions without losing context.
Models, modes, and quotas: fine-tune performance
A practical tip: combine the model selection with the Planning/Fast mode according to your needs. Planning is for in-depth research, major changes, and integration. multi-part; Fast comes from cinema for renowned, isolated Bash commands and localized tasks.
Remember that the use of premium models is subject to fees, especially in Public Preview. If a limit warning appears, adjust the model or toggle the mode. to make the most of each call.
Target audience for Antigravity
Antigravity covers everything from purely technical profiles to product roles. Front-end, back-end, and full-stack developers They will be the main beneficiaries, but also product managers and designers for rapid prototyping, and data analysts/scientists that require scripts and visualizations with deep reasoning.
Its combination of agent management, classic editing, and browser verification allows each profile to focus on its value layer. AI handles the technical execution and validation, while you retain the architecture, the decision, and the supervision.
Documentation and capabilities of the Antigravity model
The IDE relies on Gemini 3 Pro and, according to the information consulted, It was presented as a Public Preview in November 2025Beyond code completion, the key lies in its agentive capabilities: Advanced reasoning, tool usage, and autonomous programming.
In addition, options such as Gemini 2.5 and an integrated “Nano Banana” type model for generate/edit imagesThe technical cocktail enables complete workflows: planning, implementation, and validation with recordings and captures. closing the quality cycle.
Languages, menus, and preferences after installing Antigravity on Linux
From the settings menus you can customize the environment: theme, shortcuts, keyboard, and location. If you want to work in Spanish, adjust the language. Go to Preferences and review the menu bar options so that the editor, administrator, and browser behave the way you like.
Remember that Antigravity is multi-window. Editor, Agent Manager and browser each have their own surfacesThis brings clarity of responsibilities and more direct navigation between building, orchestrating, and validating.
Best practices when installing Antigravity on Linux
Initially, we recommend focusing on the flow of Agent-Assisted Development. It gives you enough autonomy without losing control. with review requests when it matters. If you're working with sensitive repositories, start with execution Off in the terminal and increase it to Auto or Turbo as you gain confidence.
Activate the Artifacts view and get used to commenting as if it were a shared document. That feedback loop is pure gold. to improve the quality and traceability of each change before mixing code.
How to try Antigravity on other platforms
Although we focus on Linux here, Antigravity is also available on Windows (x64 and ARM64) and macOS (Intel and Apple Silicon). The onboarding flow is very similarDownload, install, sign in with Google, and configure policies and model according to your needs.
The managed browser experience and artifact production is consistent across platforms, so switching between Linux, Windows, and macOS It will not force you to change the way you work.
Quick ideas to keep practicing
If you want to continue exploring, try tasks that combine agent + browser + editor. For example, it automates the extraction of data from a dashboard with charts.It requests visual validation with screenshots and generates diffs of the code that processes that data.
Another good practice is to generate living documentation: request a README tailored to your project, with steps for installation, execution, testing, and how to extend it. The agent can keep that file up to date. as you evolve the functionalities.
Antigravity takes you from "one-off prompting" to continuous collaboration with quality evidence. Between the Agent Manager, the Editor, and the integrated browserThe platform allows you to build, validate, and refine without leaving the flow, something that traditional IDEs did not solve with such elegance.