Antigravity Model Unavailable Errors Followed by Account Sign-In Failure

Google Antigravity
,
1

Description

I am experiencing a recurring issue with Google Antigravity where the selected AI model initially works correctly but becomes unavailable while executing a task. The application displays the message:

“Model unavailable, retrying in X seconds (attempt X/9).”

The task may run successfully for some time before this error appears. Once it starts, the model repeatedly retries but does not recover reliably, interrupting the task before completion.

Restarting Antigravity does not resolve the issue. The application opens normally, but the same model availability problem persists. Eventually, Antigravity unexpectedly logs me out of my Google account. When I attempt to sign in again using Google, account setup fails with the following error:

“There was an unexpected issue setting up your account.”

The accompanying technical error reads:

Post "https://daily-cloudcode-pa.googleapis.com/v1internal:loadCodeAssist": EOF

This prevents me from signing back in and continuing my work.

Steps to Reproduce

  1. Open Antigravity and sign in using a Google account.

  2. Select an AI model and assign it a task.

  3. Allow the model to begin processing the task.

  4. Observe whether the error “Model unavailable, retrying in X seconds” appears during execution.

  5. Close and restart Antigravity to check whether the problem is resolved.

  6. Observe that the model availability issue may persist.

  7. If the application logs out, attempt to sign in again using Google.

  8. Observe the account setup failure and the loadCodeAssist: EOF error.

Expected Behavior

The selected model should remain available throughout task execution, or recover automatically from temporary service interruptions. Restarting the application should restore normal operation where appropriate, and Google sign-in should continue working without unexpected account setup failures.

Actual Behavior

Model requests fail repeatedly during task execution. Restarting the application does not reliably fix the problem, and the application eventually logs me out. Subsequent Google sign-in attempts fail with an account setup error and a loadCodeAssist: EOF message.

Impact

This issue interrupts long-running tasks, wastes time, and risks losing progress. The subsequent sign-in failure prevents me from accessing Antigravity and continuing my work.

Additional Information

  • The problem has occurred during AI task execution.

  • Restarting Antigravity has not reliably resolved it.

  • The sign-in error references daily-cloudcode-pa.googleapis.com/v1internal:loadCodeAssist.

  • I have screenshots documenting the model retry loop, sign-in screen, and account setup error.

    image
    image813×502 9.99 KB

    image
    image660×484 24.4 KB

2

Hi @Elius,

Thank you for sharing the details.

Please try the following steps to resolve this:

  1. Update Antigravity App to v2.24.0: Make sure you are on the latest release (v2.24.0 in the Changelog), which adds an Authentication expired prompt with a direct Log back in button and fixes an issue where the model list could remain empty after signing out and signing back in.
  2. Test Direct Connectivity to the Service Endpoint: Run the following command in your terminal to check if your network, firewall, or ISP is resetting the TLS handshake to daily-cloudcode-pa.googleapis.com:
    • curl -sv https://daily-cloudcode-pa.googleapis.com -o /dev/null
  3. Disable Post-Quantum TLS (tlsmlkem) as a Test: Some routers, firewalls, and security software drop larger Go Post-Quantum TLS handshake packets, resulting in an immediate EOF. Try launching Antigravity with the environment variable GODEBUG=tlsmlkem=0 set:
    • macOS / Linux: export GODEBUG=tlsmlkem=0 (then launch Antigravity from the same terminal)
    • Windows (PowerShell): $env:GODEBUG="tlsmlkem=0" (then launch Antigravity from the same PowerShell window)
  4. Check VPN, Proxy, or Firewall Rules: If you are using a VPN, proxy, or local firewall/antivirus web shield, temporarily disable it (or test on a mobile hotspot) and try signing in again.

If the EOF error continues after trying these steps, please share your OS version and the output of the curl command above so we can take a closer look.

3

This is frustrating. Happens multiple times a day in antigravity 2.0 (version 2.24 even). happens with Argon, 3.8, 3.7, even 3.6. I’ve yet to experience this on antigravity cli, however.

4

Hi @formigone,

The fact that you are seeing this in Antigravity App v2.24.0 while Antigravity CLI (agy) works normally on the same machine points to desktop-specific connection and model-switching issues that were fixed after v2.24.0, along with retired model IDs if an older conversation still has 3.6 or 3.7 selected:

  1. Please update the Antigravity App from v2.24.0 to the current latest release, v2.28.1 (Download, Changelog). Specifically, since v2.24.0, Antigravity App v2.26.0 (Changelog — v2.26.0) fixed an issue where socket disconnect errors stopped the agent instead of retrying automatically, and v2.28.0 (Changelog — v2.28.0) improved performance while responses stream in and fixed an issue where a conversation could switch to a different model after Antigravity restarted or reconnected.
  2. Because Gemini 3.6 Flash and Gemini 3.7 Flash have been retired on consumer plans in favor of Gemini 3.8 Flash and Gemini 4 Argon (Choose a model), make sure to start a new conversation and select Gemini 3.8 Flash or Gemini 4 Argon rather than resuming an older thread that still has Gemini 3.6 Flash or Gemini 3.7 Flash pinned.

If you still see `“Model unavailable, retrying…”` in a new conversation after updating to **Antigravity App v2.28.1**, please reply here with a screenshot of your **Settings → Models & Usage** screen so we can look into it further.