Laya playground
a site, three games and an agent skill for an open decision model
Laya is an open-source decision model by Nandakishor M of Convai Innovations: typed questions in, real probabilities out, one forward pass, no generated text. I did not train, design or fund it. I found it while waiting on a closed competitor's waitlist, and its only marketing was a Reddit post. This is the marketing: a landing page where the model plays three games live, Tetris among them, a playground for typed questions, a 500-example benchmark against TypeSafe AI's hosted Jev with the numbers left the way they fell (Jev is more accurate out of the box; Laya is faster from a laptop, free and open), and a SKILL.md that teaches a coding agent to add the model to a project properly. The deployed site has no model behind it and replays recorded runs; the public repository runs the same pages against the real thing.
Toward publishing
100%Published
Goal
Give a good open model the front door it never had, and make brain function collapse findable as the author of the playground, the Laya vs Jev benchmark and above all the agent skill.
Stack
- html
- css
- javascript
- canvas
- python
Assets
