We write down how it works.
NiNI Labs studies the computing systems that serve AI workloads. Engineering notes explain the systems we have already built and run, and what surprised us on the way.
From the team.
- Nicrron Engineering
Prompts side by side, each on its own branch
In Ninoo Builder, several people can now prompt the same project at once. Each prompt after the first runs on its own git branch with its own live preview, and finished branches merge back into the live app, with the model combining changes that touch the same lines.
- Nicrron Engineering
Version history that only moves forward
Every change to a Builder project, by the model, by hand or by a shell command, is now a commit in a git repository only Ninoo writes to. Undo, restore and revert all add new commits, so nothing in a project's history is ever rewritten.
- Nicrron Engineering
Our own git server for Ninoo projects
Every Ninoo Builder project now has a private repository on Ninoo Git, a git server we run ourselves. People sign in with their Ninoo account, clone and push from their own computer, and pushes come back into the live app.
- Nicrron Engineering
One endpoint, many model makers
The Nicrron API puts every major model behind one OpenAI-compatible endpoint. Routing names such as nicrron/auto turn into ordered fallback chains, each model's endpoints are balanced by price and health, and repeat requests can be answered from a per-account cache at no charge.
NiNI Labs.
Applied research on inference scheduling, serverless agent runtimes, memory for long-lived agents and efficiency per token. NiNI Cloud is its first product.