After update app on MacOS, it always got the same error when I ask everything

Resolved 💬 5 comments Opened May 22, 2026 by duy12i1i7 Closed Jun 29, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.519.31651 (3017)

What subscription do you have?

GPT Plus

What platform is your computer?

Darwin 25.2.0 arm64 arm

What issue are you seeing?

it always displays like
Error running remote compact task: stream disconnected before completion: unknown variant auto, expected high or original at line 1 column 1779845

What steps can reproduce the bug?

ask everything on latest version of MacOS

What is the expected behavior?

Plz fix it

Additional information

_No response_

View original on GitHub ↗

5 Comments

github-actions[bot] contributor · 3 months ago

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

  • #24014

Powered by Codex Action

Afum · 3 months ago

I am getting the same error on Windows, in VSCode. sent a feedback we'll see: 019d2812-0367-7900-b309-22d1e82fa92d

chuygalaz · 3 months ago

I'm seeing the same issue on Windows with Codex Desktop/App:

Error running remote compact task: stream disconnected before completion: unknown variant `auto`, expected `high` or `original` at line 1 column 201651

Environment/context:

  • Platform: Windows / PowerShell
  • Codex CLI is not available in PATH (codex --version returns command not recognized), so this appears to be from the Desktop/App or IDE integration state.
  • The error persists even after editing %USERPROFILE%\.codex\config.toml to avoid model_reasoning_effort = "auto".
  • Clearing %USERPROFILE%\.codex\.tmp did not fix it.
  • Moving the newest session file did not fix it.
  • Restoring/clearing %USERPROFILE%\.codex\.codex-global-state.json affects the visible project list, but the compact error returns when the state is restored.

A targeted search under %USERPROFILE%\.codex excluding .tmp and SQLite files found matches in:

%USERPROFILE%\.codex\.codex-global-state.json
%USERPROFILE%\.codex\models_cache.json
%USERPROFILE%\.codex\sessions\...\*.jsonl

This makes Codex effectively unusable because every new prompt hits the same remote compact task failure. It looks like some persisted state/model-cache/session data can store or resend variant: auto, while the remote compact endpoint expects only high or original.

Expected behavior:

  • Codex should either migrate/normalize stale auto values in persisted state/cache, or the compact endpoint should accept/handle auto safely.
  • A corrupted/legacy session should not break all future chats/projects.
GraceHanJiarui · 3 months ago

I’m seeing the same issue in VS Code Codex with a long-running conversation thread.

Error message:

Context automatically compacted

Error running remote compact task: stream disconnected before completion: unknown variant `auto`, expected `high` or `original` at line 1 column 136131

Context:

  • Product: Codex in VS Code
  • The affected conversation is a long-running thread that I have been using for many days.
  • Today I was simply continuing the same existing conversation normally.
  • After Codex attempted Context automatically compacted, the compact task failed with the error above.
  • If I open a new VS Code window and re-enter the same conversation, I can continue chatting for a few more messages.
  • However, after a few messages, Codex attempts context compaction again and the same error appears again.
  • Other long-running conversation threads that I have also used for many days do not seem to have this issue so far.

Expected behavior:

The context compaction should either complete successfully or fail gracefully without repeatedly blocking the conversation thread.

Actual behavior:

The affected long-running conversation repeatedly fails whenever automatic context compaction is triggered, even after reopening the same conversation in a new VS Code window.

Additional notes:

This seems thread-specific rather than affecting all my Codex conversations, because my other long-running threads have not triggered this same failure yet. The error mentions unknown variant auto, expected high or original, so it may be related to how the compact task is configured for this particular thread or context state.

etraut-openai contributor · 1 month ago

Thanks for reporting this problem. Until recently, Codex used a separate endpoint and server-side logic for compaction. We recently switched to a more robust approach where the compaction logic is moved locally using the same endpoint as normal turns. This eliminates "remote compaction" errors and increases the stability and reliability of compaction operations. If you see further problems with compaction, please use /feedback and open a new bug report.