Just coming on here to make it clear that, agy --p silently drops stdout when run with a non-TTY (pipe, subprocess, redirect). Not sure if this is intended, but agy -p has no way of capturing output in non-interactive contexts. Going back to gemini-cli until this is fixed, or confirmed as intended.
Hello Team;
Is this problem fixed ?
Hey team(FEEDBACK),
After the v2.0.0 update, the IDE completely stopped working for me (chat, completions, and agents are dead). I spent today digging into the local processes and found the root cause: the release binary is pointing to your internal staging server instead of production.
Here is a breakdown of the issue and the sanitized logs.
The Core Issue
The language_server.exe process is being launched with the --cloud_code_endpoint flag set to https://daily-cloudcode-pa.googleapis.com instead of the standard cloudcode-pa.googleapis.com.
Because it is hitting the daily/staging endpoint with production OAuth tokens, Google’s backend rejects the connection and force-closes the TCP socket.
From main.log (around line 67):
Plaintext
Spawning: ...language_server.exe --standalone --override_ide_name antigravity
--override_ide_version 2.0.0
--cloud_code_endpoint https://daily-cloudcode-pa.googleapis.com
--enable_sidecars
Downstream Impact & Network Failures
The server keeps trying to hit this staging endpoint every ~25-30 seconds. Here is the sequence from language_server.log:
Plaintext
13:10:29 WaitForReady failed: context deadline exceeded
13:11:36 Cache(loadCodeAssistResponse): failed
Cache(userInfo): failed
13:14:49 wsarecv: An existing connection was forcibly closed by the remote host.
13:15:29 Brief attempt at the PRODUCTION endpoint (cloudcode-pa.googleapis.com)
13:15:32 Immediately switched BACK to staging (daily-cloudcode-pa)
13:15:33 wsarecv again on fetchUserInfo
The wsarecv error occurs because the staging server kills the connection. Since the initial auth fails, loadCodeAssist, fetchUserInfo, and fetchAvailableModels all fail downstream.
Environment & State (Sanitized)
-
OS: Windows 11 x64
-
Version: Antigravity v2.0.0
-
Process:
language_server.exe(Hovering around 223 MB private memory) -
Network: 6 active TCP connections going to
daily-cloudcode-paover IPv6, all resulting in connection resets.
The Fix / Workaround
This appears to be fixed in v2.0.1 (released May 19), but the auto-updater is failing for some users (mine downloaded the update twice to %LOCALAPPDATA%\antigravity-updater\pending\ but never actually applied it).
For anyone else stuck: manually download v2.0.1 and reinstall to clear the daily- flag.
*Note: I have full memory dumps, TCP captures, and thread dumps available if the dev team needs them for triage. I’ve excluded raw IPs, tokens, and system paths here for privacy, but let me know if you need the raw files sent securely.
OR could also try running these script in powershell :
*
- Forcefully kill the buggy IDE and the runaway Go background binary
Write-Host “Stopping all hanging Antigravity processes…” -ForegroundColor Cyan
Get-Process Antigravity, language_server -ErrorAction SilentlyContinue | Stop-Process -Force
2. Obliterate the 700+ junk XML temp files generated by the failed retry loops
Write-Host “Purging junk XML artifacts from the polling loops…” -ForegroundColor Cyan
Get-ChildItem -Path “$env:LOCALAPPDATA\Temp” -Filter “xml_file (*).xml” -ErrorAction SilentlyContinue | Remove-Item -Force
3. Clear out the broken, stale update downloader cache
Write-Host “Clearing stuck auto-updater cache…” -ForegroundColor Cyan
if (Test-Path “$env:LOCALAPPDATA\antigravity-updater”) {
Remove-Item -Recurse -Force “$env:LOCALAPPDATA\antigravity-updater*” -ErrorAction SilentlyContinue
}
Write-Host “Cleanup complete! Your system is ready for a clean manual install of v2.0.1.” -ForegroundColor Green
Why this is needed for affected users:
-
Stops the Leak: Even if you close the IDE window,
language_server.exeoften stays active in the background as a zombie process, continuously chewing up private memory. -
Reclaims Disk Space: The script safely targets only the specifically patterned trace files (
xml_file (*).xml) inside the system Temp directory without messing with any other app data. -
Resets the Updater: Wiping the
pendingfolder clears the corrupted download states so the app doesn’t loop trying to apply a broken patch.
I have the same problem. I need a solution.
Please add Antigravity 2.0 (Desktop) and Antigravity CLI to the tools.
Viewing the diff directly within the confirmation box interface (ASK) before accepting is how it should work for a truly smooth and secure experience. It would be an excellent feature.
I have ASK configured before the AGENT is going to perform an EDIT; the options appear in the interface. This should be clickable in the UI to display the changes the agent will make in Diffplex.
Please pass this feedback on to the extension/application development team so they consider adding a dropdown diff preview within the confirmation modal itself in future updates.
any response? I too am having same lockout
Did you test them out, if you did which one you can recommend among Codex, Cursor & Claude? Seems like i will also need to switch to a better develpoment enviroment.
Subject: Appeal against account/domain block – Request for review
Important: I am unable to submit this appeal through the built-in system dashboard because my account appears to be locked/restricted by your automated security filters. Every submission attempt results in an authorization error.
Dear Antigravity Support Team,
I am writing to formally appeal the restriction/blocking of my account/domain.
I understand that my business involves modeling and the 18+ industry. However, I would like to clarify that our activity on your platform strictly adheres to acceptable use policies and does not involve the distribution of adult content (pornography).
Our usage of the platform was limited exclusively to:
-
Text Captions: Mild, flirtatious texts used for social media engagement, with absolutely no explicit or pornographic materials.
-
Data Analytics: Analytical processing of numbers and statistics to optimize marketing performance.
-
Vibe Coding / Case Processing: Technical structuring, organizing, and managing business case data.
Additionally, I am experiencing a critical authorization error. When attempting to log in, the system fails to authenticate my credentials, completely blocking my access to the dashboard.
Given that no explicit content was hosted, processed, or generated, we believe this restriction may have been triggered automatically or by mistake. We kindly ask you to manually review our case and restore access.
Thank you for your time and understanding. I look forward to your response.
Title: [Bug Report] “Ghost Consumption” of Gemini Credits in Antigravity IDE – Quota Draining Automatically
Body:
Hi everyone / Support Team,
I am experiencing a severe issue with the Antigravity IDE where my Gemini model credits are continuously decreasing, even when I am not actively using the AI or sending any prompts.
It appears to be a “ghost consumption” issue. The quota drops significantly while the IDE is simply open or running in the background. I suspect this might be caused by uncontrolled background tasks, such as silent workspace indexing or runaway sub-agents, but there is no clear indicator or warning in the UI about this background consumption.
Here are the details of the issue:
-
Issue: Credits/Quota draining without active user prompts.
-
Expected Behavior: Credits should only be deducted when an active prompt or explicit AI task is triggered by the user. Background indexing should be controllable or paused automatically.
-
Impact: Rapid depletion of my available Gemini credits, resulting in wasted quota and unfair billing.
Environment details:
-
IDE Version:2.1.1
-
OS: Windows 11
-
Model Used: Gemini Pro
Could the development team please look into this bug urgently? I would also like to know if there is a process to appeal and restore the credits that were incorrectly consumed by these background processes.
Thank you for your time and assistance!
What is the best way to receive support regarding Google Antigravity?
Went to Google Antigravity - Support , but could not find which “community” is refered to? (No link, or I’m blind..)
Is it this forum that we are on? ![]()
I’m a member on three different Discord,
Google Labs, Google Gemini, Google Developer Community.
I have asked there, and I’ve asked through the “Provide Feedback” option in Antigravity Settings. Even DM’d someone on Reddit. Is the only support we get community based?
When we will see Fable class models available in Antigravity?
we want opus 5, sonnet 5, fable 5 like frontier model in antigravity.
Hello Google, Being an Ultra Plan user for 3 months now, I can finally admit I am giving up on Antigravity, had the worst experience working with it. Thank you Google.
Renewed my ultra plan yesterday, made some few changes to my codebase, and was hit with a 3hrs wait for the Opus Model, only to return and after 3 exhaustive hours of wait, to be hit with the 6days wait, without making a single change to my codebase.
Probably, not recommending anyone to buy the ultra plan or even use Antigravity at all.
It was … while it lasted, switching to Claude Code
Hello all! I’d like to bring your attention to a new topic I’ve created:
This is a way to reduce verbosity in Flash, I’ve coded it to work on 3.8, but it’s trivial to extend. Normally, I do not try to bring attention to this topic, but given that it’s a known issue that frustrates many developers (myself included), I feel like it would be beneficial to share with as many people as I can.
This reduces token usage on the agent’s side by 3/4th. It’s designed to match parity with 3.1 Pro, and this curbs 3.8 Flash’s “paranoid android” behavior. Please take a look, it might be of great value!

