About OSAI-Models
OSAI-Models is a registry for open-weight models and datasets, built by DeepNeurons AI LLC around a question the public hubs do not answer: can this checkpoint be called from a regulated transaction core, and under which integration pattern?
What we do
Public hubs tell you what a model scores. They do not tell you whether your platform can run it, whether its licence survives contact with your legal team, or what happens the first time it influences a payment.
OSAI-Models screens every repository for a static computation graph, publishes a readiness verdict and a recommended integration pattern beside the weights, derives the licence from the licence text rather than the author’s declaration, and ships the interface contract — the COBOL copybook — as part of the model record.
Who it is for
ML engineers who need to know whether a checkpoint is deployable before spending a quarter on it. Architects deciding how a model reaches a transaction core. Buyers who need deployment topology and governance posture. Risk and audit teams who need evidence before a model influences an outcome.
How we work
Every figure on the site carries its source and the date it was observed, or it is labelled illustrative. Screening verdicts are reproducible from a repository URL. Upstream metadata is stored as a claim, not a fact, and contradictions are surfaced rather than resolved quietly.
The registry mirrors what it ingests, so a repository that is relicensed or deleted upstream keeps its history here.
Status
The registry is in active development. Mainframe integration is proven on a developer workstation — export, numerical parity, an in-process call from COBOL — and awaits latency proof on a test LPAR. Figures marked illustrative are not measurements.
Contact
DeepNeurons AI LLC, [registered address]. General enquiries: [admin@deepneurons.ai]. For an enterprise evaluation, use the form on the Enterprise page.