Authoring NVIDIA-labs Object Oriented Agents (NOOA)

An agent is a Python class; its methods are its capabilities. Three rules drive everything:

• ... body = agentic method. An async method whose body ends in ... is implemented by an LLM at call time. A real body = plain Python.
• Docstring = prompt. The method docstring (plus name, signature, and class docstring) is the instruction the LLM receives.
• Return type = contract. The framework validates the LLM's output against the annotation and retries on failure.

The class docstring is part of the static system-prompt prefix — it appears in every LLM call on that agent. Keep it to a sentence or two describing the agent's role and constraints.

import asyncio
from typing import Annotated
from nooa import Agent
from nooa.unifiedllm import get_llm_client

llm = get_llm_client("gpt-4o-mini")    # any litellm model name, or a registry alias