Repository navigation
[Bug]: Model error messages render as broken raw JSON in the UI #3747
Description
Before submitting
- I searched existing issues and did not find a duplicate.
- I included enough detail to reproduce or investigate the problem.
Area
apps/desktop
Steps to reproduce
Instead of rendering as a normal error message, it shows up in a weird broken layout, with some of the message appearing duplicated or cut off.
- Open T3 Code on Arch Linux with Qtile.
- Trigger a model-related error in the composer/chat UI.
- Observe how the error is displayed.
Expected behavior
T3 Code renders model error messages incorrectly as raw JSON / broken text instead of a readable error banner. The error text appears visually garbled and partially clipped in the UI.
A clean, readable error message should appear in the UI, with proper formatting and no raw JSON leaking into the chat/composer area.
Actual behavior
The error is shown as raw JSON-like text and appears broken/clipped, for example:
{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account."}}
This seems like a UI rendering / layout issue rather than the API error itself. The error payload is being displayed directly in the interface and the container sizing/overflow looks wrong on this setup. I attached a screenshot showing the exact rendering issue.
Impact
Cosmetic issue
Version or commit
0.0.28
Environment
Arch Linux with Qtile WM.
Screenshots, recordings, or supporting files
Workaround
No response
Activity
sguergachi commentedon Jul 11, 2026
I see the same truncated error banner.
devjoca commentedon Jul 11, 2026
Reproduced on macOS desktop as well, not just Arch/Qtile.
In my case the underlying model/runtime error is a Codex + Azure Foundry failure, but the UI symptom matches this issue exactly: the full JSON error exists in app state, while the composer/chat area only shows a clipped fragment like
{"er/{...with the rest cut off.Full underlying error:
{
"error": {
"message": "X-OpenAI-Internal-Codex-Responses-Lite only supports function tools, custom tools, and client-executed tool search.",
"type": "invalid_request_error",
"param": "tools",
"code": "unsupported_value"
}
}
So this issue does not seem Linux-specific; it also affects macOS when T3 Code surfaces structured model errors.
maria-rcks commentedon Oct 9, 2026
Note
This is an automated response from a bot acting on maria's behalf.
This issue has had no activity since 2026-07-11.
Is it still relevant on the current version? If so, your version and OS would help.
juliusmarminge commentedon Oct 19, 2026
Note
Grok responding on behalf of Julius.
This should be fixed now that #21483 has landed on main. When a structured error carries a readable
error.message(like the examples here), the collapsed chat banner now shows that message instead of raw JSON, and the full original payload is still available behind "Show full error". The earlier clipped/broken layout was addressed in #3899.Closing this. If you still see raw JSON or a clipped banner on a build that includes #21483, please open a new issue with the error text and your version/OS.
fix(server): a terminal that loses the workspace-recovery race opens …