System OneSystem One

NOULXP · THE OPEN STANDARD

One standard for every decision model.

NoulXP is an open standard for packaging and serving decision models: models that answer typed questions (a choice, a score, or a noul, the probability that a statement is true) with calibrated confidence. Package a model once, and it runs on any machine and answers every application and AI agent the same way.

What it gives you

One interfaceEvery NoulXP model takes the same request, a state and typed questions, and returns the same answer, with a calibrated probability for every option. Swap one model for another and your code stays the same.
Runs anywherepip install noulxp, then run a package on a CPU, an NVIDIA GPU or Apple hardware. ONNX weights run through ONNX Runtime (noulxp[onnx]), GGUF weights through llama.cpp (noulxp[gguf]).
Safe to runA package holds weights and data, not code. Its manifest, noulxp.json, lists every file with its SHA-256. Templates are filled in and the weights run through ONNX Runtime or llama.cpp: nothing in a package is executed.
The model’s own answersEach package carries a conformance file: what the model’s own code answers on a fixed set of requests. noulxp check replays it, and passes only when every decision is the same and every probability is within 0.01.

The tools

One Python package, noulxp, is the reference implementation, for Python 3.11 or newer. Its commands take a model from its checkpoint to a server:

noulxp exportConvert a model’s checkpoint into a package. Converters exist for Laya, Julia 1, Decider and AnyJev.
noulxp checkReplay the package’s conformance file on your machine: the same decision on every question, or exactly where an answer differs.
noulxp benchDecisions per second, latency and the cost per 1,000 decisions, for a package or for any server that speaks NoulXP.
noulxp calibrateFit how sure the model says it is to your own labelled requests, without touching its weights.
noulxp serveAnswer requests over HTTP on any machine, in the same format as the System One inference API.
noulxp mcpGive packages to AI agents as a tool over the Model Context Protocol: Claude, Cursor and other MCP clients.

Publish it on System One Models

Publish a NoulXP package with your model here. When the package passes its check, the model earns the NoulXP compatible badge, on its page, on its card in every list and on your profile, and it is listed with the compatible models. Ask for a live playground, and once it is approved the System One Engine serves the model: a playground on its page, and a hosted API your users call with a key.

Earn the badge Get started

Open source

The specification, its JSON Schemas, the converters and the reference runtime are licensed Apache-2.0, and anyone may build an engine for NoulXP. Converted packages carry the models’ own weights, which stay under each model’s licence.