NVIDIA Releases NOOA: An Object-Oriented Python Framework That Turns an AI Agent Into a Single Class

NVIDIA Labs released NOOA (NVIDIA Object-Oriented Agents), an open-source, model-agnostic Python framework that simplifies AI agent development by collapsing everything into one Python class. In this design, methods are the actions the model can take, fields are agent state, docstrings serve as prompts, and type annotations are enforced contracts. A method whose body is '...' is completed at runtime by an LLM-driven loop, while a normal method body remains deterministic Python that the model can call as a tool. The framework ships with two strategies: PredictStrategy, a single typed LLM call with a retry loop, and CodeActStrategy, an iterative Python REPL where the model calls execute_python until it submits return_result, which is validated. NVIDIA reports capability tests achieving 97.9% pass rate across 88 tests run five times on ten models, and end-to-end results of 82.2% on SWE-bench Verified with GPT-5.5, 86.8% on CyberGym L1, and 85.1% mean RHAE on ARC-AGI-3 with GPT-5.6-sol. The framework uses a pass-by-reference mechanism to keep large data live in the REPL, avoiding context compaction. NOOA is Apache 2.0, installable via pip (v0.0.8), requires Python 3.12–3.13, and is marked as an alpha research preview. Because agents can execute LLM-generated code, NVIDIA advises running them only inside OS-level isolation like a container, VM, or OpenShell. Models can be plugged in via LiteLLM, supporting hosted APIs, Ollama, and vLLM endpoints.