Getting Started with NVIDIA NOOA Python Agents

NVIDIA NOOA (labs for object-oriented Python agents) is an open teaching/lab codebase for class-as-agent design—not a mass-market “install and chat” product. Repo: github.com/NVIDIA-NeMo/labs-OO-Agents.

For after-work learners it is an advanced pattern lab: you read how agents are structured as objects, then steal ideas into smaller personal projects.

Who this is for

After-work Python learners who already know classes/__init__/methods and have run at least one simpler agent framework (smolagents, Pydantic AI, etc.). Absolute beginners should park this until Intro + one lighter agent post are done.

Overview: Top 10 AI agent tools for after-work Python.

Steps to start tonight

1. Clone the lab repo (read-only mindset)

git clone https://github.com/NVIDIA-NeMo/labs-OO-Agents.git
cd labs-OO-Agents

Skim the README for Python version, extras, and which notebook/script is the “hello” path. Labs move—trust the README over blogs.

2. Create a venv and install what README lists

python -m venv .venv
source .venv/bin/activate
# pip install -r requirements.txt   # or whatever the README specifies

3. Configure credentials only if the sample needs a live model

Some cells may be offline-readable. If a live LLM is required, use env vars. Never commit .env.

4. Run one class-as-agent example end-to-end

Execute the shortest example that constructs an agent class and handles one user task. Print the object’s attributes/methods you care about (plan, act, tools, etc.—names vary).

5. Write a design postcard in NOTES.md

Answer in five lines: What state lives on the instance? What is a method vs a free function? What would you delete for a personal mini-port?

Tiny exercise (20–30 minutes)

Without rewriting the whole lab, add one method or tool stub (e.g. ping() -> str returning "pong") and call it from the existing loop if the design allows. If the architecture fights you, document why in NOTES—that observation is the exercise.

Pitfalls

• Expecting a polished consumer CLI. This is a lab/OO teaching repo.
• Running every notebook on night one. One path.
• Skipping class fundamentals. If self is fuzzy, do Intro first.
• Copy-pasting GPU-cluster assumptions onto a laptop. Follow local/CPU notes in README.
• Shipping lab code to production as-is. Learn patterns; harden elsewhere.

Why this fits after work

OO agent design is easier to study in a quiet evening than during standup noise—and patterns transfer to whatever framework you use at work later.

Next step

If class design still feels heavy, strengthen foundations with Intro to Python Programming before your next framework tour.