What NILS is

NILS keeps imaging data as one registry: every file is read once, every scan is classified by a rule pack, every person is a pseudonym, every question is a stored document with a hash, and every release is described before it is made.

It is three programs, and you install only the ones you need.

Program	What it is	Needs
the engine (nils)	The registry itself, and the one door to it: a command line and an HTTP API.	nothing
the desk (nils-desk)	The web application over one engine, and the place people sign in.	an engine
the assistant	Turns a sentence into a question the engine answers.	an engine, a model gateway, a model

The engine alone is a product

One person on one laptop can digest a tree of DICOM files, classify it, ask questions, and write a release out, without installing anything else. The desk exists so that other people can do the same without a terminal. The assistant exists so that they can do it in words.

That order is also the order to install in. Get the engine working first; everything else talks to it.

What a registry is

A directory (or a Postgres schema) that holds what the engine learned from your files: the subjects, their sessions, the stacks and what each one is, the classifications with their evidence, the jobs that ran, and the audit of who did what. It never holds a copy of your DICOM files: it points at them where they are. Files that arrive identified are pseudonymised into a tree of their own first, and the registry reads only that tree.

The key in the registry is what pseudonyms are derived from. Two registries built with the same key give a subject the same code; with different keys, different codes. It is the one secret that matters, and it is yours.

Licence

The engine, the desk and the assistant are AGPL-3.0-only; the contracts between them are Apache-2.0. The source is at github.com/kineuro/nils, nils-desk, nils-assistant and kvasir.