NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework designed to streamline AI agent development. Traditionally, building an agent involves juggling prompt templates, tool schemas, callback hooks, and workflow graphs. NOOA consolidates these elements into a single Python class, where methods define the actions an agent can take, fields represent agent state, docstrings serve as prompts, and type annotations enforce runtime contracts. A method with an ellipsis (...) body is completed at runtime by an LLM-driven loop, while methods with normal bodies execute as deterministic Python. This unified interface allows developers and models to share a common ground, making agent behavior testable, traceable, refactorable, and version-controlled like any conventional software.
Performance and Efficiency
In benchmarking, NOOA achieves impressive results: 82.2% on SWE-bench Verified, 86.8% on CyberGym L1, and a mean RHAE of 85.1% on ARC-AGI-3. Notably, it accomplishes this while using roughly half the tokens consumed by the open harnesses it was compared against, highlighting its efficiency.
Deployment and Safety
NOOA is deployable, but with a critical caveat: it requires OS-level isolation. The framework is released under the Apache 2.0 license and can be installed via pip install nooa (v0.0.8, released July 30, 2026). It requires Python 3.12–3.13. On PyPI, it is classified as alpha status, and NVIDIA describes it as a research preview. Since agents can execute LLM-generated code, NVIDIA explicitly states that its AST checks and module deny-lists are defense-in-depth guardrails—not a containment boundary. The actual containment boundary must be a container, a VM, or NVIDIA OpenShell. Models are pluggable through LiteLLM, supporting hosted APIs, Ollama, and vLLM endpoints.
As the AI landscape evolves into 2026, NOOA represents a significant step toward making agent development more accessible and robust, aligning with the industry's push for modular, maintainable AI systems.
via MarkTechPost