Agent reasoning should respect user's language

🌍 Supporting this important i18n/UX issue

This is a critical user experience problem that affects all non-English speaking users of Codex CLI worldwide. I want to highlight why this deserves attention from the team.

Related Issues (showing this is a recurring community need)

This issue is directly related to several previously reported problems:

• #7143 - "The thought process display does not adhere to the language specified in the global AGENTS file" (closed)
• - #6321 - "Getting some foreign language in Codex CLI and output contains Thinking content and stopped in between" (closed)
• - - #5804 - "/review command ignores the AGENTS.md"

The fact that this keeps being reported shows real user demand for proper language support.

Why This Matters

User Demographics: Codex has a global user base. Users from Ukraine, Russia, China, Japan, Germany, France, Brazil, and many other countries communicate naturally in their native languages. When reasoning appears in English while the user writes in their native language, it creates:

1. Cognitive friction — users must context-switch between languages
2. 2. Lost transparency — the reasoning trace loses its value if users can't easily understand it
3. 3. Inconsistent experience — especially when compared to Claude Code which handles this gracefully

Proposed Implementation

The reasoning_language = "auto" config option suggested in this issue is elegant and backward-compatible. It could:

• Default to "auto" (detect from user input)
• - Allow explicit override: "en", "uk", "ru", "zh", "ja", etc.
• - - Require minimal model changes since the underlying models already support multilingual output

Call to Action

If you've experienced this issue, please:

• 👍 React to the original issue
• - Share your experience in the comments
• - - Mention which language you use with Codex

cc @etraut-openai (who responded to similar issues before) — would love to hear if this is on the roadmap!

---
Supporting better UX for the global developer community 🚀

Thanks for the feature request. We generally prioritize features based on community feedback (upvotes). We haven't (yet) received much demand for this feature, but we will leave it open and see if it receives more upvotes.

@etraut-openai Thank you for the thoughtful response and for keeping this issue open!

I completely understand the prioritization approach based on community upvotes. A few thoughts that might be helpful:

Visibility challenge: Non-English speaking users are often less active in English-language issue trackers, which may lead to underrepresentation of their needs. The fact that related issues (#7143, #6321, #5804) keep appearing suggests latent demand that doesn't always translate into upvotes.

Low-hanging fruit: Since the underlying models already support multilingual output, this feature could potentially be implemented with minimal changes — primarily an instruction modification or a simple config option.

I'll do my part by sharing this issue in developer communities where Codex is discussed in other languages. Hopefully, this will help surface the actual level of interest.

Thank you again for your time and consideration! 🙏

Thanks for the feature request. We generally prioritize features based on community feedback (upvotes). We haven't (yet) received much demand for this feature, but we will leave it open and see if it receives more upvotes.

Curious if this has been addressed or not already, I personally am not affected or have not noticed it at all but seems like it currently has the most total reactions for the Codex CLI (tag) in issues here and has been stale it seems this year.

If it has not been addressed I would be willing to see if I can PR something. Just let me know, thanks!