Skip to content

[Bug]: Antigravity sign-in succeeds but session/new fails with "Internal error"; models never load (Windows) #27761

Closed

Description

@Iansung8

Environment

  • T3 Code (Beta) desktop 0.1.5; the provider environment is the local machine
  • Windows 11 Pro 25H2 (OS build 26200.8037), x64
  • Antigravity runtime v1.4.1 (installed through T3; reinstalling did not help)

Actual behavior

Settings > Providers > Antigravity shows "Not authenticated" and "Antigravity authenticated, but could not initialize a session or load models." Retry sign-in fails the same way every time.

In server.trace.ndjson, AntigravityAuth.runSignIn completes the Google sign-in, but the session/new RPC fails after about 3 seconds:

session/new  exit: Failure  cause: "Error: Internal error at AntigravityAuth.runSignIn (binCli-1k0PQRR7.mjs:192705:24)"

After that, refreshing models fails with "Could not refresh Antigravity models. The previous model list is unchanged."

This is not the slow-startup timeout or an HTTP 429. No 429 appears in the logs, and the failure comes in about 3 seconds. On the same machine, with the same Google account, the Antigravity CLI streams responses from streamGenerateContent normally.

Expected behavior

session/new succeeds and the model list loads.

Notes

The trace does not record the agent's stderr text, so the underlying error is not visible. Happy to provide more logs if you tell me where they are.

Activity

  1. 1DogeGamer commented on Oct 31, 2026

    @1DogeGamer

    Additional report from a Windows x64 desktop (OS build 26300), using T3 Code 0.1.5 with the managed Antigravity runtime 1.4.1.

    After Google sign-in and submission of the full 127.0.0.1 redirect URL, T3 shows “Antigravity authenticated, but could not initialize a session or load models.” The local server.trace.ndjson records session/new failures at 2026-10-31 08:06:06Z and 08:06:33Z, with Error: Internal error in AntigravityAuth.runSignIn. The user reports standalone Antigravity sign-in works.

    The .gemini/config/skills target exists, and both T3-managed skills junctions resolve to it, so the dangling-junction workaround in #27976 does not explain this report. The trace does not expose the underlying runtime error. No OAuth URL or token values are included.

  2. Iansung8 commented on Oct 31, 2026

    @Iansung8
    Author

    Update: after upgrading to T3 Code v0.1.6 (stable, same Windows 11 machine, Antigravity runtime 1.4.1), Antigravity sign-in now works. Status shows "Authenticated · Google account" and the model list loads with no errors. Probably fixed by one of the Antigravity changes in v0.1.6 (possibly #27153). Feel free to close if no one else still sees it.

  3. t3dotgg commented on Nov 1, 2026

    @t3dotgg
    Member

    Note

    🤖 GPT-6.1 Terra [1m] responding on behalf of Theo

    Closing as resolved after the reporter verified Google sign-in and model loading on stable 0.1.6 on the same Windows machine and Antigravity runtime. #27761 comment #27761 comment

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions