›_ Codex Issues
Codex Issues / #40312

[Windows] Signing out shows “Oops, an error has occurred” after AppServer RPC disconnect

Open 💬 3 comments Opened Aug 24, 2026 by Pete2000
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Summary

Signing out of the Codex desktop app completes successfully, but the app does not transition to the normal signed-out screen. Instead, it immediately reaches the global error boundary and displays:

Oops, an error has occurred

The application log reports:

AppServerManager RPC is not connected

Environment

  • App: OpenAI Codex desktop app
  • Version: 26.818.5229.0
  • Installation source: Microsoft Store
  • Package status: OK
  • OS: Windows 25H2, OS build 26200.9168
  • Architecture: x64

Microsoft Store currently reports that this is the latest version available for my release channel.

Steps to reproduce

  1. Open the Codex desktop app.
  2. Sign in with a ChatGPT account.
  3. Open the account menu/settings.
  4. Select Sign out.
  5. Wait for the app to switch to the signed-out screen.

Expected behavior

After signing out, the app should display the normal signed-out screen with the option to continue and sign in again.

Actual behavior

The sign-out operation succeeds, but the UI displays a generic error page:

Oops, an error has occurred

A Try again button is shown.

Reproduction frequency

This has reproduced consistently during recent sign-out attempts. Local logs indicate that the same AppServerManager RPC is not connected error has occurred immediately after multiple sign-out operations since approximately August 19, 2026.

Sanitized log excerpt

The following timestamps are in UTC:

2026-08-24T01:43:22.615Z
AppServerConnection response_routed
method=account/logout
errorCode=null

2026-08-24T01:43:23.127Z
desktop_fetch_auth_401
hadToken=false
skipRetryReason=no_token_attached

2026-08-24T01:43:23.225Z
error boundary
errorMessage="AppServerManager RPC is not connected"
errorName=Error

View original on GitHub ↗

3 Comments

github-actions[bot] contributor · 4 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #40293
  • #40197
  • #39208

Powered by Codex Action

kenjikanaiwagj-creator · 18 hours ago

Additional Windows reproduction on newer desktop builds

I can confirm that this still reproduces on the current Windows desktop build 26.820.9563.0 (x64), as well as on 26.820.7780.0.

Environment:

  • Windows 11 25H2, build 26200.9168, Professional
  • Locale: ja-JP
  • Installation: Microsoft Store package

Sanitized event sequence on 26.820.9563.0 (UTC):

2026-08-27T16:38:43.678Z INFO  account/logout completed durationMs=735 errorCode=null
2026-08-27T16:38:43.904Z WARN  request HTTP 401 hadToken=false skipRetryReason=no_token_attached
2026-08-27T16:38:44.009Z ERROR renderer error boundary: AppServerManager RPC is not connected
2026-08-27T16:38:51.961Z ERROR layered config load failed: AppServerManager RPC is not connected
2026-08-27T16:38:59.726Z WARN  /wham/usage HTTP 401 hadToken=false
2026-08-27T16:39:59.832Z WARN  /wham/usage HTTP 401 hadToken=false
2026-08-27T16:40:59.912Z WARN  /wham/usage HTTP 401 hadToken=false
2026-08-27T16:41:59.981Z WARN  /wham/usage HTTP 401 hadToken=false

The logout request succeeds and the token is cleared, but the renderer reaches the RPC error boundary 331 ms after the logout response. The background process then continues unauthenticated usage polling until the Codex process tree is terminated.

The earlier reproduction on build 26.820.7780.0 had the same ordering, with the RPC error 326 ms after the successful logout response and unauthenticated polling continuing for about 30 minutes.

This supports an ordering/race issue in the logout transition: the App Server RPC lifecycle appears to be torn down before all renderer consumers have switched to the signed-out state.

Workaround: wait for active tasks to finish, terminate the Codex desktop process tree, relaunch the app, and sign in again. No cache or configuration reset is required.

The excerpt above contains no account IDs, conversation IDs, request IDs, access tokens, usernames, local paths, or full stack traces.

kirillsaven · 1 hour ago

Additional Windows evidence: successful logout followed by cleanup-time RPC access

Independent confirmation on Windows 11 Pro x64 (build 26200), official Desktop 26.820.9563.0. Two owner-initiated logout attempts on August 27 produced this ordering:

  1. account/logout completed successfully.
  2. Approximately 938 ms and 473 ms later, respectively, the renderer entered the error boundary with: AppServerManager RPC is not connected.
  3. The backend remained alive until the app was fully exited through its tray menu. Relaunching then allowed sign-in.

This is not, by itself, evidence of a backend crash or SQLite corruption.

Read-only inspection of the packaged frontend found a plausible cleanup path: the review-pane snapshot-metrics component calls an RPC accessor after its AppServer manager has disconnected. In the affected bundle, webview/assets/app-initial-BHiyAZSX.js, the component/accessor symbols are Irl / qg. The same accessor/cleanup pattern was present in the separately inspected 26.820.10647.0 package. These are version-specific observations and a candidate cause, not a maintainer-confirmed diagnosis.

Could the signed-out transition detach/disable RPC consumers before disconnecting the manager, and make unmount cleanup tolerate an already-disconnected manager? A regression should verify successful logout reaches the sign-in view without an error boundary, while preserving local history.

The machine is now on 26.825.3734.0, but logout has not been deliberately reproduced on that build, so I am not claiming current-build reproduction or a verified fix. A separate history-projection recovery succeeded; that does not establish that this logout bug is fixed.

No signed application files were patched. No account/session IDs, private paths, tokens, raw logs, or conversation contents are included.