NVIDIA Open-Sources NOOA: Build an AI Agent as a Single Python Class

Summary: NVIDIA open-sourced NOOA, a framework where one Python class is the entire agent — fields hold state, methods expose capabilities, docstrings serve as prompts, and type annotations define model contracts.

Key Points

• Core idea: one Python class per agent — no separate orchestration layer, scaffolding framework, or prompt templates required
• Benchmark claim: 82.2% on SWE-bench Verified (GPT-5.5 backend), beating prior state-of-the-art while using half the tokens and eliminating context compaction
• Availability: Apache 2.0, pip install nooa (v0.0.8), Python 3.12–3.13; currently classified as alpha / research preview
• Security caveat: NVIDIA explicitly warns that NOOA executes LLM-generated Python, which could exfiltrate data or delete files — "not a containment boundary"

Why It Matters

NOOA challenges the assumption that agent quality lives in the model alone — its benchmark numbers suggest the harness around the model may matter just as much. Paired with the launch of a 37-member Open Secure AI Alliance, NVIDIA is positioning itself as the infrastructure layer for the next wave of agentic software. The alpha label warrants caution; independent benchmark reproduction is needed.

Read More

• NVIDIA NOOA overview (MarkTechPost) — MarkTechPost
• Nvidia's NOOA: one class, one agent (The New Stack) — The New Stack