NVIDIA AI Releases NOOA Python Framework for AI Agents

NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework designed to streamline the development of AI agents. The framework consolidates disparate components of agent development, such as prompt templates, tool schemas, callback code, and workflow graphs, into a single Python class structure. Within this class, methods represent the actions an AI agent can perform, fields store its state, and docstrings function as prompts. Type annotations are used to enforce contracts at runtime. Methods can be designed to be completed by an LLM-driven loop at runtime, or they can contain deterministic Python code. This unified interface allows developers and models to share a common ground, enabling agent behavior to be tested, traced, refactored, and version-controlled using standard software development practices. NVIDIA reports that NOOA achieved 82.2% on the SWE-bench Verified benchmark, 86.8% on CyberGym L1, and an average of 85.1% RHAE on ARC-AGI-3. Notably, these results were obtained using approximately half the tokens compared to the open harnesses against which it was benchmarked. NOOA is deployable, but it requires OS-level isolation for execution. The framework is licensed under Apache 2.0 and can be installed via pip as version 0.0.8, which was released on July 30, 2026. It is compatible with Python versions 3.12 and 3.13. PyPI classifies NOOA as an alpha release, and NVIDIA characterizes it as a research preview. The framework supports agents that can execute LLM-generated code. NVIDIA explicitly states that its Abstract Syntax Tree (AST) checks and module deny-lists serve as defense-in-depth guardrails rather than absolute containment boundaries. The actual containment is achieved through external mechanisms such as containers, virtual machines, or NVIDIA OpenShell. NOOA's pluggable model architecture supports various LLM providers through LiteLLM, allowing integration with hosted APIs, Ollama, and vLLM endpoints. The framework is targeted at AI-native startups and mid-market platform teams building internal agents, as well as enterprise AI platform and applied-research groups conducting evaluations or pilot projects. However, NVIDIA advises that regulated production workloads should await a stable release. Industries that can benefit from NOOA include developer tooling, cybersecurity, cloud and DevOps, data analytics, financial services operations, and customer support. Potential applications range from repository issue triage and patching to terminal and infrastructure automation, vulnerability validation pipelines, and large-batch classification tasks.