NVIDIA Object Oriented Agents
A Python AI agent framework that supports local Ollama and vLLM models through LiteLLM, with typed methods, generated Python actions and execution traces.
NVIDIA Object Oriented Agents (NOOA) is a Python framework for developers building AI agents in their own projects. It supports local models through Ollama and vLLM via LiteLLM, with Qwen3 examples for both backends. It also supports Anthropic and OpenAI models, which send requests to those remote providers.
NOOA keeps an agent's state, prompts and capabilities together in a Python class. Developers can combine ordinary deterministic methods with methods whose behavior comes from an LLM, while type annotations define the expected inputs and results. Agents can pass live Python objects between methods and compose other agents as objects.
Models can act by generating and executing Python with access to the agent's objects and helper methods. This approach reduces the need for separate tool definitions and lets developers work with familiar Python testing and refactoring practices. Typed results include automatic retries.
A browser trace viewer shows model calls, executed code and method calls, including how one operation relates to another. Tracing requires the viewer to be running. Optional packages add long-term memory, evaluation tools and a coding agent for Agent Client Protocol editors such as Zed.
NOOA is research software. Generated code can delete files or send private data elsewhere; its code checks don't contain those actions. Agents that execute generated code need isolation at the operating system level, such as a container, virtual machine or NVIDIA OpenShell.