# \[Bug\]: Managed ChatGPT conversations keep failing after token renewal until T3 restarts · Issue \#16132 · pingdotgg/t3code

[View on GitCafe](https://git.cafe/pingdotgg/t3code/issues/16132)

Repository: [pingdotgg/t3code](https://git.cafe/pingdotgg/t3code)

Visibility: public

State: completed

Priority: 0

## Description

[Discussion and activity](https://git.cafe/pingdotgg/t3code/issues/16132)

---

### Before submitting

- [x] I searched existing issues and did not find an exact duplicate.
- [x] I included enough detail to reproduce or investigate the problem.

### Area

apps/server — managed Codex with ChatGPT sign-in, observed in the desktop app.

### Steps to reproduce

This is the sequence observed on the affected machine, rather than a fresh controlled reproduction:

1. Sign in to ChatGPT through T3's managed Codex setup.
2. Use a conversation successfully and keep T3 running.
3. After the access token expires, send another message. It fails with `401 token_expired`.
4. Refresh or toggle the provider in Settings. In the captured incident, T3 successfully renewed the credentials, but conversation requests continued failing.
5. Restart T3 and continue the conversation. Requests work again without another sign-in.

The same problem has not been observed on a second machine using an existing Codex CLI. The two setups are compared below.

### Expected behavior

After T3 renews the ChatGPT credentials, conversations continue working without an application restart or another sign-in.

### Actual behavior

The saved credentials renew successfully, but conversation requests continue failing with `401 token_expired`. Restarting T3 restores access without signing in again.

This interrupts ongoing work even though the ChatGPT connection can still be renewed.

### Impact

Major degradation or frequent failure. Conversations on the affected setup stop working and require an application restart to recover.

### Version or commit

T3 Code `0.0.46-nightly.20261005.2676` on both machines.
Build commit: `7812230572f2`.

### Environment

| | Affected setup | Comparison setup |
|---|---|---|
| OS | Windows | Linux |
| Codex setup | T3-managed Codex with ChatGPT sign-in | Existing Codex CLI |
| Codex version | `0.156.1` | `0.160.0` |
| Observed behavior | Conversations fail after token expiration | This problem has not been observed |

The authentication setup appears to be the relevant difference. This comparison does not establish that the issue is Windows-specific; the Codex versions also differ.

### Logs or stack traces

The reported error is `401 token_expired`.

The captured sequence on October 5, 2026, in UTC:

- **16:22:** Conversation requests fail with the expired-token error.
- **16:28:** T3 successfully renews the credentials.
- **16:29–16:30:** Conversation requests still fail with the same error.
- **16:31:** After restarting T3, a conversation produces a response without another sign-in.

### Screenshots, recordings, or supporting files

[Supporting investigation and evidence](https://gist.github.com/chrhicks/4ba9f9a3edc0e7ac0b599644e98ba657).

The separate material contains the sanitized logs, source analysis, independent test, and its limitations.

### Workaround

Restart T3. Toggling the provider has sometimes helped, but did not restore conversations in the captured incident.

### Related issues

- #14356 concerns the refresh request itself failing with `invalid_grant`. Here, renewal succeeds and conversation requests continue failing.
- #11136 describes stale authentication after switching CLI accounts. This report concerns automatic renewal with T3-managed ChatGPT sign-in.