> I need an AI harness that lets me switch LLM models dynamically depending on the task.

OpenRouter put up something about this a few days ago. Check out their Advisor and Subagent docs.