Could you still utilize this tool running a local model (any Ollama model for example) instead of making calls out to Anthropic to avoid costs while running this tool or does it have something specific to Anthropic that requires it?