When you spend 1k/month of LLM api calls. That doesn’t sound all too unreasonable to have locally in the near future.