What version of the IDE extension are you using?
openai.chatgpt 0.4.79 (also observed previously on 0.5.79)
What subscription do you have?
ChatGPT Business (member)
Which IDE are you using?
VS Code 1.109.5 (072586267e68ece9a47aa43f8c108e0dcbf44622, x64)
What platform is your computer?
Windows 11 Pro 25H2 (build 26200.7840)
What issue are you seeing?
Codex in VS Code does not stream assistant output progressively. Messages are often rendered only at the end, or UI updates are delayed until user interaction (click/focus/change in UI).
Reproduces in a fully pristine VS Code profile with only the Codex extension installed.
Actual behavior:
- Streaming is effectively broken in UI (message appears in bulk / at end).
- Output -> Codex warning flood:
- Received broadcast but no handler is configured method=thread-stream-state-changed
- Repeated errors:
- local-environments is not supported in the extension
Key log evidence (from attached bundle, logs/Codex.log):
- thread-stream-state-changed: 193
- client-status-changed: 1
- local-environments is not supported: 2
- startup websocket prewarm setup failed: 0
- stream disconnected - retrying sampling request: 0
- falling back to HTTP: 0
What steps can reproduce the bug?
Pristine profile repro:
- Launch VS Code with a new empty --user-data-dir and --extensions-dir.
- Install only openai.chatgpt@0.4.79.
- Sign in with ChatGPT login (no API key).
- Open Codex chat and ask for a longer response (e.g., “List numbers 1..200, each on a new line”).
- Observe that output is not rendered token-by-token in the UI.
What is the expected behavior?
- Token-by-token streaming in the Codex webview without requiring user interaction.
- No unhandled broadcast spam for stream state changes.
Additional information
What was already tried:
- VS Code reload/restart.
- Single window only (no WSL/Remote-SSH windows open).
- Workspace storage reset.
- Extension reinstall/downgrade/upgrade (0.4.79 / 0.5.79).
- Clean profile with isolated CODEX_HOME and isolated extensions dir.
- Disabling potential local interference (e.g., VPN service process) did not resolve.
Regression signal:
- User reports issue started after first usage with additional Remote-SSH window(s).
- After that point, issue persisted even in local-only and pristine profile scenarios.
Attachments:
- codex-streaming-diagnostics-safe-20260303-155648.zip (logs/Codex.log, logs/exthost.log, HIGHLIGHTS.txt, extension-metadata/package.json)
codex-streaming-diagnostics-safe-20260303-155648.zip
What version of the IDE extension are you using?
openai.chatgpt 0.4.79 (also observed previously on 0.5.79)
What subscription do you have?
ChatGPT Business (member)
Which IDE are you using?
VS Code 1.109.5 (072586267e68ece9a47aa43f8c108e0dcbf44622, x64)
What platform is your computer?
Windows 11 Pro 25H2 (build 26200.7840)
What issue are you seeing?
Codex in VS Code does not stream assistant output progressively. Messages are often rendered only at the end, or UI updates are delayed until user interaction (click/focus/change in UI).
Reproduces in a fully pristine VS Code profile with only the Codex extension installed.
Actual behavior:
Key log evidence (from attached bundle, logs/Codex.log):
What steps can reproduce the bug?
Pristine profile repro:
What is the expected behavior?
Additional information
What was already tried:
Regression signal:
Attachments:
codex-streaming-diagnostics-safe-20260303-155648.zip