Codex CLI sessions become unrecoverable after ~37-40 minutes on Windows 11 after websocket disconnects

[kfawell]

Title: Codex CLI sessions become unrecoverable after ~37-40 minutes on Windows 11 after websocket disconnects

What version of Codex CLI is running?

codex-cli 0.121.0

Which model were you using?

gpt-5.4

What platform is your computer?

Windows 11

What terminal emulator are you using?

PowerShell

Network environment

Home network, no intentional VPN/proxy in use.

What issue are you seeing?

In Codex CLI, a session works normally for a while, then starts failing with websocket disconnect/reconnect behavior. Once a given session enters this state, it no longer recovers. Opening a brand new session works again temporarily, but after roughly the same amount of time, that new session also starts failing.

The affected session is not just interrupted mid-turn once; it becomes effectively poisoned:

• re-opening the same session is possible
• but as soon as I send another prompt, it spends a long time trying to connect/reconnect
• it does not resume normal operation
• starting a fresh session is the only workaround

User-visible error

Stream disconnected before completion: failed to send websocket request: IO error: An existing connection was forcibly closed by the remote host. (os error 10054)

What steps can reproduce the bug?

1. Start a new Codex CLI session on Windows 11.
2. Use the session normally for around 30-40 minutes.
3. Continue sending prompts after that point.
4. Eventually the session starts hitting websocket disconnect/retry behavior.
5. After that happens, the same session no longer recovers; any new prompt in that session triggers long reconnect attempts.
6. Start a fresh session.
7. The fresh session works again, but later fails the same way.

What is the expected behavior?

If a websocket transport drops mid-session, the session should either:

• reconnect cleanly and continue working, or
• fail once for that turn but remain usable for future prompts.

A session should not become permanently unusable after one of these disconnects.

What do you see instead?

The session retries repeatedly and then remains stuck in a reconnect/failure loop. A new session works, but the old session appears permanently unrecoverable.

Timing observed locally

I checked the local session metadata and codex-tui.log timestamps. Two recent CLI sessions on the same machine showed:

• Session A:

  • started: 2026-04-19T20:38:04.277Z
  • first disconnect: 2026-04-19T21:18:06.307Z
  • elapsed: about 40m 02s

• Session B:

  • started: 2026-04-20T18:52:37.394Z
  • first disconnect: 2026-04-20T19:29:51.868Z
  • elapsed: about 37m 14s

So the first failure appears after roughly 37-40 minutes in recent sessions.

Relevant sanitized log excerpts

2026-04-19T21:18:06.307294Z  WARN ... thread.id=<redacted> ... codex_core::codex:
stream disconnected - retrying sampling request (1/5 in 213ms)...

2026-04-19T21:29:44.689782Z  INFO ... thread.id=<redacted> ... codex_core::codex:
Turn error: stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses)

2026-04-20T19:29:51.868205Z  WARN ... thread.id=<redacted> ... codex_core::codex:
stream disconnected - retrying sampling request (1/5 in 208ms)...

2026-04-20T19:31:35.011644Z  WARN ... thread.id=<redacted> ... codex_core::client:
falling back to HTTP

2026-04-20T19:33:13.257185Z  WARN ... thread.id=<redacted> ... codex_core::codex:
stream disconnected - retrying sampling request (1/5 in 196ms)...

Additional notes

This does not look like a one-off total network outage because:

• new sessions can still be started after the failure
• those new sessions work normally for a while
• the problem is that older affected sessions do not recover once they start disconnecting

[etraut-openai]

If you can repro, please use /feedback to upload logs and session details, then post the thread ID here.

[Mahkhmood9]

@kfawell repro = (meaning reproduce ... if you can reproduce/recreate)