- Officially available for macOS, Windows, and Linux with a Google account.
- Autonomous agents with integrated browser and verifiable artifacts.
- Configurable Planning/Fast modes and execution and review policies.
If you've come here looking to download Google Antigravity for Windows and Mac , this is the guide you've been waiting for: what exactly this IDE with autonomous agents is, how to install it step by step on macOS and Windows, requirements, initial setup, usage tips, and solutions to common errors. All explained in a practical way, using everyday language, so you don't get lost in technical jargon.
Antigravity is a development platform that puts AI agents at the center : they plan, program, test, browse the web, and validate their own work while you focus on architecture and decision-making. Throughout this article, you'll see how to leverage Mission Control (the agent manager), the VS Code-like editor, the integrated browser , and the verifiable artifacts that document everything the agents do.
What is Google Antigravity and why does it matter?
Antigravity is an agent-first IDE that evolves the classic editing experience into an orchestration of autonomous tasks. Instead of simply auto-completing, agents can plan and execute complex engineering tasks asynchronously, even working in parallel on different objectives while you monitor progress from a mission control-style dashboard.
The interface is supported by three surfaces: an Agent Manager to orchestrate goals, a VS Code-based Editor to edit and review changes, and deep integration with Chrome that allows agents to test applications as a person would (clicks, forms, navigation, captures and recordings).
In addition to leveraging models like the Gemini 3 Pro , the environment presents results through "artifacts"—clear deliverables that justify decisions and changes (plans, diffs, screenshots and videos , test results). This closes the infamous "trust gap": it's no longer a matter of "trust me," but rather "look at the evidence."
According to available information, the launch of Antigravity has come in parallel with advances in Gemini 3 and performance benchmarks such as SWE-bench Verified and Terminal-Bench , positioning it as a very serious proposition against assistants that only complete code.
Availability, requirements, and Google account
Antigravity is available for macOS, Windows, and Linux . The documentation and guides list two sets of requirements: one mentions support for macOS 10.15 and later, 64-bit Windows 10/11, and popular Linux distributions; the other specifies optimization for Apple Silicon on Mac, x64/ARM64 support on Windows, and glibc/glibcxx compatibility on Linux. If your hardware is limited, you should review both lists before installing.
You'll need a Google account (the guides recommend a personal Gmail account in the preview) to authenticate and use the templates. During startup, an associated Chrome profile is created to enable agent-assisted navigation, so don't be surprised to see file, network, and accessibility permissions on the system.
A helpful tip: it's been mentioned that the preview includes free quotas for premium models. If you see limit warnings, you may have used up your daily quota for the top-of-the-range model; the app itself will let you know.
Download and install on macOS
For Mac, the process is straightforward: go to antigravity.google/download , download the installer for your architecture (recent guides suggest "Download for Apple Silicon"), and open the .dmg file . Drag the app to Applications, launch it, and complete the Google sign-in. On the first run, Antigravity initializes the workspace and downloads components, which may take a few minutes.
macOS may display the classic "unverified developer" warning on newly released apps. If this happens, enable " Open Anyway " in System Settings > Privacy & Security, and check accessibility or disk permissions if the system prompts you for them.
During the initial setup wizard, you can import settings from VS Code or Cursor, choose a theme (dark or light), and define the agent's operating mode. It's best to follow the recommended path and refine the details later: Antigravity is flexible and allows for on-the-fly changes.
Download and install on Windows
On Windows, go to antigravity.google/download and choose the x64 or ARM64 installer depending on your computer. Double-click the .exe file and follow the wizard (if SmartScreen warns you, click "More info" > "Run anyway"). You can leave the default path as "C:\Program Files\Google\Antigravity".
When finished, open the application from the Start menu, authenticate with your Google account , and wait for it to sync components. If you use corporate policies, your IT department may need to approve the use of OAuth with Antigravity; in that case, request the appropriate permission.
As in macOS, the first boot activates the wizard to import configurations, adjust the theme, and define the extent to which the agent will have autonomy when executing commands or requesting revisions.
Linux installation (and quick notes)
On Linux, there are two ways to install: using a repository (APT or DNF, depending on the distribution) or directly downloading .deb/.tar.gz packages. The most common examples are Ubuntu 20+, Debian 10+, Fedora, and RHEL, which require glibc >= 2.28 and glibcxx >= 3.4.25. If you encounter broken dependencies, running ` apt update && apt upgrade` before installation can save you a lot of trouble.
For Fedora and derivatives, DNF usually only requires a simple ` sudo dnf install antigravity` once the package is in the appropriate repositories. After installation, launch `antigravity` from your app launcher or terminal and complete the Google login process.
Initial Assistant: Agent Autonomy and Policies
Once inside, you'll see a wizard where you define the agent's workflow. Antigravity offers four preset approaches: Agent-Driven Development , Agent-Assisted Development , Revision-Based Development , and Custom Configuration . The "Assisted" mode is usually the most balanced middle ground.
This setting is complemented by two fine-grained policies. The Terminal Execution Policy determines whether the agent can execute shell commands without requesting permission: “Off” (never except on a whitelist), “Auto” (decides when to request approval), and “Turbo” (executes except on a blocklist). You decide the level of trust.
The second is the Review Policy : “Always Proceed” (no review requested), “Agent Decides” (the agent decides whether to consult), and “Request Review” (always requests your approval). These options balance human control and speed, and can be changed during the conversation.
Mission Control (Agent Manager) and Editor
Antigravity clearly divides the experience between the Agent Manager and the Editor. When you open the app, you'll typically see Mission Control: there you define high-level goals ("refactor authentication", "update dependencies", "generate billing proofs") and monitor the progress of multiple agents in parallel.
Mission Control features the Inbox with all your conversations, "Start Conversation" to open a new one, "Workspaces" to switch between local folders, "Playground" as a drafts area, and shortcuts to switch to Editor view or the browser, all within easy reach.
In each conversation you can choose the model (depending on availability) and the agent mode: “Planning” if you need in-depth research, detailed plans and artifacts, or “Fast” if you urgently need a small change (renaming, simple commands, specific tweaks).
The Editor, for its part, retains the familiarity of the VS Code environment (file explorer, syntax, extensions) and adds the agent panel on the right and "inline commands" to select code and request contextual changes without leaving the file.
When the agent needs to interact with the web, it invokes a browser sub-agent that operates in a Chrome browser managed by Antigravity. This sub-agent can read the DOM, scroll, click, type, capture screenshots, log the console, and even record video of its session.
The first time the agent tries to "open the website," the app will guide you through installing the Chrome extension . The typical flow: you start a conversation (for example, in Playground), ask "go to antigravity.google," the agent detects missing configuration and shows you the "Configure" button. You install the extension and grant permissions.
Once it's ready, you'll see that the managed browser is framed with a blue border when the agent controls it. From Mission Control, you can grant specific permissions and see in real time which pages it accesses and what artifacts it generates after its work.
Verifiable artifacts: evidence of what AI does
The agent's "language" with you is artifacts . It's not limited to a chat; it provides you with verifiable pieces: task lists and plans, code diffs, "before/after" screenshots, browser recordings , and unit or integration test results.
In the Admin view, you'll see a toggle to show/hide artifacts. If you turn it on, a list of evidence associated with the current conversation will appear: you can review them, leave comments similar to Google Docs, and request iterations on a specific point.
This approach provides transparency: when the agent claims to have "fixed a bug," you review the tangible evidence (diffs, passing tests, video of the workflow). Ideal for teams that want traceability and for reviewing without having to run everything locally.
Practical use cases that work
To get a better idea, three real-world examples are very helpful. First, a simple news-gathering task : ask the agent to visit Google News, extract headlines, and summarize the relevant information for you. You'll see how it opens the browser, navigates, captures data, and returns the results.
Second, the creation of a dynamic website with Python and Flask for a one-day event (eight talks, speakers, search by category/author/title, lunch break, simulated data). The agent plans, implements, starts a local server, and provides you with the URL; if port 5000 is busy, it will automatically look for alternatives (e.g., 8080).
Third, a productivity app with a Pomodoro timer . The agent generates the front-end, validates states in the embedded browser, and produces a media artifact with the verification video. If you request style changes or want to add a timer image, update the plan, iterate, and test again.
Finally, here's a test example: starting with a Python module (order logic with inventory, payment processing, discounts, and exceptions), the agent can generate unit tests with mocks , run them, and attach the reports. You keep the test code and the execution evidence, ready for CI.
Tricks to get the most out of agents
Specific instructions perform better: instead of "log in," request a JWT login , password recovery via email, Redis sessions, and attempt limits. The more contextual information, the more refined the plan the agent generates.
Divide large projects into phases : backend APIs first, then frontend, and finally integrations. This gives you checkpoints to review artifacts, request changes, and keep the architecture aligned.
Share artifacts in team code reviews . They help understand why specific decisions were made and facilitate knowledge transfer, which is key in distributed teams.
Common problems and how to overcome them
On macOS, you might see the message “ Unable to verify developer .” Solution: System Settings > Privacy & Security > Allow to open anyway. This is the standard procedure for new apps.
If Windows Defender blocks the installation, use " More info " > "Run anyway" and, if necessary, add an exclusion from Windows Security. Also, always make sure to download from the official website.
On Linux, if dependencies fail, check that you meet the glibc/glibcxx requirements and perform a system upgrade before installing the Antigravity package. On older distributions, you may need to update your version.
With Google Sign-In, some users have reported that browser authentication works fine, but the link to the app is slow or doesn't work at all on the first try. There have been cases where users tried Brave, Chrome, Safari, or Zen, and after several attempts (in one case, a dozen with Brave and a personal account), it finally worked. If you experience this, clear your cache/cookies and try a different browser and account combination.
Security, data and control in companies
As with any cloud-based AI tool, it's advisable to review Antigravity's data policy : what data is stored, for how long, and whether it's used for training. Many companies require opt-outs for training with their proprietary code.
For extremely sensitive components (key algorithms or security-critical modules ), it may be prudent to maintain local development and use Antigravity for less confidential tasks while the legal and security team evaluates the workflow.
If you use Google Workspace, enable access controls and auditing. It's important that IT can manage permissions, monitor usage, and enforce policies in line with compliance requirements.
Features that make the difference
Antigravity isn't tied to a single model: in addition to Gemini 3 , options with Claude Sonnet 4.5 and GPT-OSS variants are mentioned. This allows you to choose the most suitable AI for each task (complex architecture, detailed documentation, or rapid prototyping).
Artifact- based communication replaces loose text with structured reports that include reasoning and trade-offs. For managers and code reviewers, this traceability reduces friction and raises the bar for quality in remote teams.
The “native browser” testing with the Chrome extension allows agents to test interfaces like QA would: they click, fill out forms, navigate, and verify real-world behaviors that static analysis would miss.
The public preview mentions a generous free plan with limits that, according to early experiences, allow users to build several complete apps per day. Once the product stabilizes, paid plans for teams and businesses are expected to be introduced.
Competition, market and positioning
The ecosystem is moving fast. Platforms like Cursor have grown significantly, and Antigravity is making waves with an agent-based approach and deeper browser integration. Key milestones in the sector, such as market valuations and acquisitions, have been cited, along with the arrival of Gemini 3 and its significant leap forward in reasoning.
Beyond the headlines, the key is that the developer shifts from "writing every line" to being an architect and validator , relying on agents to handle the mechanical aspects. It's not about replacing talent, it's about multiplying it with a different workflow.
Learning resources and community
To get comfortable with it, you have the official documentation at antigravity.google/docs , video tutorials (including a "learn the basics" tutorial with an engineer from the team), the blog for news, and an active community on Reddit/Discord with examples of prompts and flows.
You'll also find repositories with sample projects (from simple websites to microservices), ideal for understanding how the agent dissects each problem and what artifacts it produces in the process.
Third-party download sources: an important distinction
While there are repositories and popular app stores, remember that Antigravity is a cross-platform desktop IDE geared towards development. For example, Uptodown presents itself as a cross-platform store specializing in Android, aiming to provide legal and unrestricted access to a large catalog; it's useful in its own right, but for Antigravity, it's recommended to download from the official Google website to avoid integrity issues or outdated versions.
If you're looking to download Google Antigravity on Windows or Mac, you already have the complete roadmap: requirements, installation, initial agent autonomy decisions, how to use Mission Control and the Editor, enabling the browser, validating with artifacts, and resolving typical issues. With this foundation, you can put agents to work planning, programming, and testing while you focus on designing the solution with sound judgment.
