Play within bounds.
Nemotron Lightning chooses controls from visible game feedback. Actions, requests, time and estimated spend have explicit limits.
Self-EvolvingVideo GameQA
The evidence workflow
Control → Observe → Review
An open workbench.
A self-evolving vision.
AutoPlay QA connects NVIDIA Nemotron to recorded game feedback, native replay and developer review. Our vision is AI testers that improve through experience.
Keep every source.
Nemotron Ultra reviews recorded evidence with step citations. Software signals, AI judgments and human review stay distinguishable.
Make it replayable.
Inspect before and after states, export reports and replay controls in the original runtime. Developers can check the claim themselves.
Inspect the play.
Question the verdict.
Explore recorded Grid and Unity sessions. Follow the controls, inspect the feedback and read the model’s cited review.Playback uses saved observations and makes no model requests.
AUTOPLAY QA WORKSPACE
Every game has a story to test.
Start a bounded playtest and turn surprising game behavior into reproducible developer evidence.
CaptureExploreReview
Better testers.
Built together.
A shared foundation for exploring how AI can investigate games. The next step is a community effort, with inspectable evidence at its core.
More game interfaces
Grid and Unity adapters form the current showcase. An experimental Mars browser adapter explores editor workflows.
Tests that start with a question
An experimental protocol records hypotheses, predictions and control experiments before actions, with regression export tooling.
Experience worth keeping
Archived context and memory prototypes provide a starting point for community research into stronger investigation and learning.