NVIDIA’s new agent framework treats agents as Python classes with state and tools
sterlingcrispin · x · 2026-07-25
NVIDIA frames agents as Python objects with state, tools, and LLM-driven methods
The post highlights a new NVIDIA framework where an agent is modeled as a single Python class:
- fields represent agent state
- functions are the actions the agent can take
- docstrings act as prompts
- ordinary deterministic Python can be mixed with non-deterministic ... blocks that the model fills in
What the example shows
The screenshot describes a support agent in which:
- normal methods run as plain, testable Python
- a Predict method makes a single typed LLM call
- a CodeAct method runs an iterative loop where the model can inspect live objects, call helper methods, and return a typed result
The author says this is interesting and plans to try it out, while wondering whether large models may eventually be RL-trained or fine-tuned with this harness.
Related event: NVIDIA Introduces Object-Oriented Python Agent Framework(2 posts)→
More from coding & agent
- MCP fault injection study: agent task scores drop 60% under realistic tool failures, scale doesn't help — dair_ai · 2026-10-26
- General agents burn ~140M tokens/month on LinkedIn outreach; scripts do it 60x cheaper — ChrisHarpon2 · 2026-10-26
- OpenAI upgrades Decisions API: 400 questions per call, native voice and image support — OpenAIDevs · 2026-10-26
- NLI alone rejects 5/6 new agent memories; a 0.45 similarity gate fixes it all — Alternative_Pin9598 · 2026-10-26
- LangChain ships Managed Deep Agents 0.9, walkthrough from Interrupt London keynote — LangChain · 2026-10-26
- Knowledge Worker Kit adds a full section on organizing projects — dSebastien · 2026-10-26