NVIDIA-NeMo/labs-OO-Agents ? reverse-engineered prompt

Build me a Python library for making AI agents in a more natural, object oriented way.

I want to define an agent as a normal Python class, where regular methods are just normal code, and methods with empty bodies get filled in by the model at runtime. The class should let me keep state on the object, use typed inputs and outputs, and write the prompt in the docstring so the method name and signature become part of the contract.

Please include a simple quick start example, a few runnable examples showing structured output, tools, tracing, and a basic command line entry point for viewing traces and running agent demos. It should work with different LLM providers, both hosted and local, and feel easy to use from a normal Python project.

If you need to check current docs or model support details, look them up online. Make it clean, testable, and friendly for people who want to build agents without wiring up a bunch of separate tool schemas and callback code.