I reproduced the issue in a Windows Clean Boot.
The game remains responsive and keeps established TCP connections to Ollama on 127.0.0.1:11434.
Ollama and the model remain fully responsive.
I enabled OLLAMA_DEBUG_LOG_REQUESTS=1 and captured the exact failing /api/chat request.
The request JSON is valid. Replaying the exact same request manually with curl returns HTTP 200 and a complete response in about 7.8 seconds with done:true and done_reason:"stop".
So the request itself is valid and Ollama can answer it. The issue appears to be in the game's handling of the Ollama response or the request lifecycle, not in system load, Ollama availability, or the model.
If you want, I can e-mail you a detailed bug report.