Interesting! What were you guys trying to achieve by running them in your own tool harness?

We had already implemented tool_calls in our own database and have a system that executes them and creates our conversation array, etc. So we wanted to leverage the providers that RubyLLM supports without having to change the tool execution in our platform.