From trained model to reviewed prediction.

MLsuite turns a trained model into a form people can use. Upload an .mlbundle, shape its inputs and reports into a versioned schema, run single or bulk predictions, ask colleagues to review the results and export inputs, outputs and answers as one CSV.

Quick start Open MLsuite Self-host it

The Predict page of a bookmark: the form's results on the left, a session of named runs waiting to be saved on the right. — A bookmark's run page: the generated form, its results and the session of runs waiting to be saved.

This is the manual for MLsuite, an open-source platform for running machine-learning models through generated forms and collecting human feedback on their results. It covers every option of the web application, the client API that outside software calls, and how to run your own instance. The screens come from the public instance at mlsuite.irlab.org, where anyone can create an account and try it.

Find your path

Make a first predictionUpload a bundle, create a schema, bookmark it and run it.Shape the formLabels, field kinds, layout and review questions in the schema JSON.Collect feedbackAssign inferences to reviewers and read their answers.Get the data outOne CSV with inputs, outputs and every reviewer’s answers.Share a model publiclyA public page anyone can run, with examples and a publication note.Call it from softwareAPI clients, mls_ keys and the prediction endpoint.

How it fits together

MLsuite keeps the whole loop in one place, and keeps every result tied to what produced it:

Step	Where	What is kept
Upload a trained model	Models	The `.mlbundle` file, its SHA-256 and what the bundle says about the model.
Describe its inputs and outputs	Schemas	Immutable, numbered snapshots of the form and its model bindings.
Run it	Predict	Each saved run, with its inputs, every model’s result and the exact snapshot.
Review the results	Review	Each reviewer’s answers, per inference and per report.
Use the data	Inferences and export	A filterable table and a CSV of everything above.

Models come from MLbundle, which packages scikit-learn, XGBoost, CatBoost and ONNX models without pickle. Forms are drawn by MLform. Both are part of eshkra, like MLsuite. MLsuite is MIT licensed; read the source on GitHub.