[Windows][Bug] Antigravity app is unusable; possible GPU/Electron/Chromium launch issue

Hi Google Antigravity team,

I would like to report a Windows issue where the Antigravity app itself became unusable.

Summary

The main issue is that Google Antigravity app no longer launches or becomes unusable on my Windows PC.

During troubleshooting, I also found that Discord, another Electron/Chromium-based desktop app, had a similar launch problem. Discord could be launched only by using the --in-process-gpu workaround.

Because of this, I suspect this may be related to GPU acceleration, Electron/Chromium rendering, GPU cache, graphics driver compatibility, or a similar Windows desktop runtime issue.

I cannot prove that Antigravity directly caused the Discord issue. However, the important point is that Antigravity itself is currently not usable on my PC.

Environment

  • OS: Windows

  • GPU: Intel UHD Graphics 630

  • Additional GPU: NVIDIA RTX 2070 SUPER

  • App affected: Google Antigravity for Windows

  • Related app observed during troubleshooting: Discord desktop app

Main problem

  • Antigravity app does not start normally / cannot be used.

  • I tried multiple troubleshooting steps, but the Antigravity app still does not work.

  • Because of this, I currently cannot use Antigravity app and may need to operate through Antigravity CLI instead.

Related observation

Discord also failed to launch normally, but it could start with this workaround:

$discordExe = Get-ChildItem "$env:LOCALAPPDATA\Discord" -Filter "Discord.exe" -Recurse -ErrorAction SilentlyContinue |
Sort-Object FullName -Descending |
Select-Object -First 1 -ExpandProperty FullName

& $discordExe --in-process-gpu

This suggests that the issue may be related to GPU acceleration or Electron/Chromium rendering on Windows.

Troubleshooting already tried

  • Restarted Windows

  • Ran Windows system file check / repair

  • Tried launching Discord with GPU workaround

  • Tried changing Windows graphics settings

  • Tried moving or clearing app cache

  • Tried assigning Discord to RTX 2070 SUPER

  • Tried similar troubleshooting ideas for Antigravity, but Antigravity app still does not launch correctly

Expected behavior

Antigravity app should launch normally on Windows and remain usable.

Request

Could the Antigravity team investigate whether the Windows version of Antigravity has known launch issues related to GPU acceleration, Electron/Chromium rendering, graphics driver compatibility, cache corruption, or multi-GPU environments?

I would also appreciate any official recovery steps for users whose Antigravity app cannot launch, especially when the app UI is blank, stuck, or fails to open.

Thank you.

Having the same problem, report below:

1. Environment Information

  • Operating System: Windows 11

  • Hardware Architecture:

    • CPU: AMD Ryzen 5 5500

    • RAM: 16 GB

    • GPU: NVIDIA GeForce RTX 4060

  • Application Context:

    • This involves Antigravity 2.0 and the Antigravity IDE.
  • Version & Update History:

    • When I first downloaded version 2.0.10 from the official website, both Antigravity 2.0 and the IDE worked perfectly. The crash-on-launch issue (with Antigravity 2.0) only began after I performed an uninstallation and reinstallation cycle.

    • While troubleshooting the broken application by running the backend server manually via PowerShell, the background auto-update successfully triggered and downloaded version 2.0.11 (though even with the update working, the application still failed to render any visual screen).

    • I subsequently located this updated 2.0.11 installer executable inside my %APPDATA%\Roaming folder, copied it over to my Desktop, and used it for my later reinstallation attempts rather than reverting to the website’s 2.0.10 package.

2. Steps to Reproduce

  1. Installed Antigravity 2.0 (the companion tool beside the IDE).

  2. Installed the Antigravity IDE on the same machine. (Note: At this stage, both tools launched and operated perfectly).

  3. Uninstalled Antigravity 2.0 via standard Windows uninstallation procedures, leaving the Antigravity IDE untouched (it was working fine until my last try to fix the Antigravity 2.0, which have broke also the IDE).

  4. Attempted to reinstall Antigravity 2.0.

3. Observed Problem

  • Upon completing the reinstallation process, the Antigravity 2.0 application window immediately crashes and closes in less than 1 second instead of opening.

  • Attempting to launch the application manually via the direct shortcut or running antigravity.exe yields the exact same behavior (immediate silent closure).

  • At this point, the problem was only with the Antigravity 2.0 and not the IDE.

4. Troubleshooting Steps, Port Analysis, and Error Logs

A. Electron GUI Core Process Failure

When I ran the frontend wrapper, the Chromium Embedded Framework failed to spin up its rendering contexts, generating a fatal exit code in my terminal:

Plaintext

[24976:0603/224704.915:ERROR:content\browser\gpu\gpu_process_host.cc:996] GPU process exited unexpectedly: exit_code=-2147483645
[24976:0603/224704.950:ERROR:content\browser\gpu\gpu_process_host.cc:996] GPU process exited unexpectedly: exit_code=-2147483645
[24976:0603/224704.979:ERROR:content\browser\gpu\gpu_process_host.cc:996] GPU process exited unexpectedly: exit_code=-2147483645
22:47:05.003 > Starting app (v2.0.10) with dynamic port…
[24976:0603/224705.083:FATAL:content\browser\gpu\gpu_data_manager_impl_private.cc:417] GPU process isn't usable. Goodbye.

B. Standalone Language Server Diagnostics

Isolating my execution by launching .\language_server.exe directly via PowerShell bypassed the immediate GUI crash and successfully spun up local listener ports, but revealed internal network anomalies and missing protocols:

Plaintext

W0603 23:01:00.528457  7208 launchmanager.go:68] Entering local chrome mode! This is WRONG unless you are running tests or in eval mode on Linux.
W0603 23:01:00.528990  7208 log_context.go:117] Cache(loadCodeAssistResponse): Singleflight refresh failed: Post "/v1internal:loadCodeAssist": unsupported protocol scheme ""
E0603 23:01:00.531098  7208 log.go:398] Failed to poll ListExperiments: Post "/v1internal:listExperiments": unsupported protocol scheme ""
E0603 23:01:00.531628  7208 log.go:398] Failed to poll FetchAvailableModels: failed to get load code assist response: Post "/v1internal:loadCodeAssist": unsupported protocol scheme ""
E0603 23:01:00.835528  7208 error_parsing.go:45] Error: Post "/v1internal:loadCodeAssist": unsupported protocol scheme ""

C. Local SSL/TLS Handshake and Network Layer Gridlock

The server listened concurrently on an HTTPS (gRPC) port and a fallback HTTP port. However, trying to network into these ports locally produced authentication and handshake failures:

D. Model Context Protocol (MCP) Instance Failures

When I manually terminated the standalone server session via Ctrl+C, the system reported a failure to close all secondary spawned tool instances cleanly:

Plaintext

I0603 23:36:36.705970 11336 server.go:2105] Got signal interrupt, shutting down
I0603 23:36:36.719807 11336 mcp_manager.go:1395] Failed to close MCP instance "github-mcp-server": exit status 1
E0603 23:36:36.719807 11336 mcp_manager.go:1395] Failed to close MCP instance "chrome-devtools-mcp": exit status 1
E0603 23:36:36.719807 11336 mcp_manager.go:1395] Failed to close MCP instance "taskmanager": exit status 1

E. Web Browser Console DOM Exceptions

Routing an external browser instance into the standalone backend server page skipped the initial silent crash but stopped at an unrendered, static black interface. Inspecting the DOM console (F12) exposed unhandled promise rejections originating from main.js:

Plaintext

[TrajectorySummaries] connecting...
main.js:11741 [AppState] connecting...

Uncaught (in promise) Error: Missing LS CSRF token cookie
    at d$ (main.js:10839:374)
    at Z8a (main.js:11898:132)

Uncaught (in promise) Error: No native storage bridge found during initialization
    at F6a.init (main.js:10869:260)
    at T6a (main.js:10881:33)
    at U6a (main.js:10882:376)
    at Z8a (main.js:11898:325)
    at async main.js:11910:87

F. Command Flag Parsing Restrictions

Attempting to force web security or origin rule overrides directly through execution flags caused a clean execution crash, indicating the following defined parameter usage bounds:

Plaintext

flags provided but not defined: -allow_all_origins -disable_csrf
Usage of language_server.exe:
  -app_data_dir="antigravity-ide"
  -csrf_token=""
  -http_server_port=0
  -https_server_port=0
  -use_ls_chrome_devtools_mcp=true

5. Remediation Actions Attempted

Action 1: Deep Environment Purge and Full Reinstallation (Failed)

As my first attempt to resolve the issue, I performed a complete manual wipe of the environment:

  • I fully uninstalled both Antigravity 2.0 and the Antigravity IDE from my system.

  • I completely deleted all local system directories under %LOCALAPPDATA% containing references, caches, or configurations for antigravity, antigravity-ide, and gemini.

  • In addition to %LOCALAPPDATA%, I also thoroughly cleaned and removed any residual files, configuration folders, or metadata associated with these three tools located directly within my Home directory (%USERPROFILE%).

  • Following this deep data purge, I attempted a fresh reinstallation. The immediate application closure behavior and underlying subprocess failures remained completely unchanged.

Action 2: Scripted Directory and Windows Registry Cleansing (Failed)

Hoping to catch any persistent state files or hidden configurations left behind by the standard uninstaller, I ran explicit PowerShell scripts to target system caches and registry hives:

  • I executed commands to forcefully delete the application data runtime environment located at %APPDATA%\Roaming\Antigravity and the updater cache at %LOCALAPPDATA%\antigravity-updater.

  • I also purged the Windows Registry keys assigned to the application lifecycle located under HKCU:\Software\antigravity.

  • After running these cleaning commands, I reinstalled the application once more. The application continued to instantly exit upon launch.

Action 3: Onboarding Reset Workflow & Resulting IDE Regression (Failed)

In a subsequent troubleshooting step, I executed the Antigravity: Reset onboarding feature through the command palette interface of the running Antigravity IDE.

This action did not resolve the application closure/crash loop. Instead, it introduced an independent system regression:the native, built-in security/code extension (identifying as SafeCode / CodeSafer or similar, i can’t remember now.) completely disappeared from my IDE environment. It is currently omitted entirely from both my “active/enabled” and “inactive/disabled” extensions index trees.