Codex CLI Assistant incorrectly decodes UTF-8 characters on Windows

[claumigue]

Description

The TablePlus Assistant appears to incorrectly decode UTF-8 output when using Codex CLI on Windows.

Unicode characters returned by Codex are displayed as mojibake in the Assistant.

For example:

Expected:

café — don’t “quote” € ñ

Displayed in the TablePlus Assistant:

café — don’t “quote� € ñ

Steps to reproduce

1. Open TablePlus on Windows.

2. Configure the LLM Agent to use Codex CLI.

3. Open the Assistant sidebar.

4. Send:

   Reply exactly with: café — don’t “quote” € ñ

5. Observe that the returned Unicode characters are corrupted.

Control test

I ran the exact same prompt directly in Codex CLI.

Codex CLI correctly outputs:

café — don’t “quote” € ñ

So the Codex CLI itself appears to be producing the UTF-8 text correctly.

Environment

• OS: Windows 10 Pro, version 22H2
• OS build: 19045.6466
• TablePlus: 26.8.0 (372)
• LLM Agent: Codex CLI
• Codex CLI: 0.147.0
• Model: gpt-5.6-terra
• Codex authentication: ChatGPT Plus
• Codex executable detected by TablePlus:
  C:\Program Files\nodejs\codex.cmd

Additional observation

The same issue occurs with normal English responses containing typographic apostrophes.

For example:

I don’t have access...

is displayed as:

I don’t have access...

This appears to be an encoding/decoding issue between the Codex CLI output and the TablePlus Assistant on Windows.

[sonswift]

Hi @claumigue, thanks for the detailed report. We’ve identified and fixed it. The fix will be included in the next TablePlus Windows update.