While an AI does have the capacity to respond in any language that it has learned, this is a bit strange.
My native language is ENGLISH, and yet it will, on its own start responding in GERMAN, SPANISH, SWEDISH, RUSSIAN. I have to literally tell it to keep all responses in English. 9 or 10 responses later, it will start responding again in another language.
might be relevant to rules/skills/codebase? perhaps if codebase has all non-english comments?
I do not think agent has access to your geolocation or system preferences, but I might be worng
Cursor has no control over this, this is being done by the underlying LLM. Hence the question whether there is anything pointing to non-english languages.
That’s what I thought. But since this was done in Agent mode WITHOUT selecting an actual agent then I thought where is this coming from, hence the quandry.
Hey, this looks more like a model behavior thing than Cursor itself.
Cursor doesn’t control output language — it’s the model that decides based on your prompt + context.
From my own experience: I’m a native Indonesian speaker but I work mostly in English. Sometimes when using GPT or Claude, if my code/comments/docs have any Indonesian text mixed in, the model starts replying in Indonesian. Mixed-language input often triggers mixed-language output.
A few things that would help diagnose:
• Which model are you using? (GPT 5.5, Claude, Codex, etc.)
• Any non-English text in your codebase, comments, file names, or rules?
• Can you share a screenshot of when it happens?
If it keeps happening with pure English input on a clean repo, that’s likely a real bug — worth submitting to the Cursor bug report forum so the team can track it.