Made sure Ollama sends a text instead of sending back JSONs of generated / used tokens and uses gpt-oss:20b