Codex Desktop dictation fails with "No active organization found"

Open 💬 2 comments Opened May 11, 2026 by tomarova-commits
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Summary

Codex Desktop dictation records audio, but transcription fails after pressing Stop.

A feedback report with session logs was already submitted from the app.

Correlation ID:

019e0140-e0a0-7042-9c57-5e525eec628a

Environment

  • App: Codex Desktop
  • Version: 26.506.31421
  • OS: macOS 15.5
  • Account type: personal account

Observed behavior

Voice dictation starts and records audio. After stopping the recording, Codex shows that the audio cannot be transcribed.

Local logs show:

[Composer] dictation failed
{"detail":"No active organization found"}

The regular chat flow still works, and account/read succeeds in the logs.

Steps already tried

  • Checked microphone access
  • Restarted Codex
  • Logged out and back into ChatGPT
  • Logged out and back into Codex
  • Fully reset local Codex app state by renaming ~/Library/Application Support/Codex
  • Tried short 2-3 second recordings

Expected behavior

Dictation should transcribe the recorded audio in the composer.

Actual behavior

Dictation fails every time with No active organization found.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 3 months ago

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

  • #22136
  • #21927

Powered by Codex Action

coldwaterkimkim · 3 months ago

Additional repro from macOS on the latest public Codex app version I can install.

Environment

  • App: Codex Desktop
  • Version: 26.513.20950 / bundle version 2816
  • CLI bundled with app: codex-cli 0.131.0-alpha.9
  • Runtime noted in app logs: Electron 42.0.1
  • OS: macOS 26.4.1 (25E253), Apple Silicon / arm64
  • Auth mode: ChatGPT login, not API-key auth
  • ChatGPT plan: Pro
  • Date/time: 2026-05-16 KST

Behavior

Dictation starts and records audio, but stopping/submitting dictation shows:

Unable to transcribe audio

Fresh Codex Desktop logs after a full app restart still show:

[Composer] dictation failed errorMessage="{"detail":"No active organization found"}"

Nearby log lines show account/read succeeds immediately before/after the dictation failure, so the app is generally signed in and usable, but the transcription backend path appears to lack an active org/workspace context.

Example from the fresh run:

2026-05-16T01:49:00.130Z error [electron-message-handler] [Composer] dictation failed errorMessage="{"detail":"No active organization found"}"
2026-05-16T01:49:02.718Z info [AppServerConnection] response_routed ... method=account/read ... errorCode=null
2026-05-16T01:49:13.674Z info [AppServerConnection] response_routed ... method=account/read ... errorCode=null
2026-05-16T01:49:13.881Z error [electron-message-handler] [Composer] dictation failed errorMessage="{"detail":"No active organization found"}"

Earlier diagnostics also showed the transcription request hitting:

POST https://chatgpt.com/backend-api/transcribe -> 400
{"detail":"No active organization found"}

Local checks / attempted fixes

  • Confirmed macOS microphone permission for com.openai.codex is allowed.
  • Confirmed default input device is the built-in MacBook microphone.
  • Confirmed ~/.codex/auth.json is auth_mode: chatgpt and does not contain an API key.
  • Re-authenticated via codex login --device-auth; auth refresh succeeded.
  • Verified codex login status reports logged in using ChatGPT.
  • Verified model catalog works after re-auth (codex debug models returns models).
  • Fully quit/reopened Codex Desktop and retried dictation.
  • Checked latest appcast: installed app 26.513.20950 appears to be the current public Codex app version.
  • Compared Codex Electron cookie stores. The main app cookie jar initially had only Cloudflare cookies for chatgpt.com, while the Codex browser partition had ChatGPT session cookies. Copying non-CF OpenAI/ChatGPT cookie rows from the browser partition into the main app cookie store did not resolve the issue after restart.

Notes

I am not attaching full raw logs here because they may include local paths/session metadata, but the fresh failure is consistently the same No active organization found error. This seems less like a local microphone/audio capture problem and more like the Desktop dictation code path is calling backend-api/transcribe without the active ChatGPT org/workspace context that the web backend expects.