Contact support
Email [email protected] with “Kwip ML” in the subject. Include your app version, macOS version, Mac model, the action you took and the exact message shown.
Start small. Describe the problem using a bundled example or a small invented dataset if possible. Do not send confidential datasets, credentials, learned weights or a full project/run archive without first discussing what is needed.
If a verified run offers a Support summary, review its contents and use Save reviewed summary… to save a local copy. You decide whether to attach it to an email; the app does not send it automatically.
Start with a working example
Open Learn, choose a journey or example, and follow its instructions on the Build canvas. Connect and configure the blocks, then run the supported flow. The bottom Results area shows reached data, run output and evaluation information; History keeps recorded runs for review.
Some lessons are guided practice; others assess a specific result. Completing a guided example is not a certification of mastery.
Source files and columns
Choose a source file through the app’s file chooser. Check the separator, header choice and preview before applying changes. If a file moved or access was revoked, select it again. In global Settings, Permissions lists saved source access.
Downstream column choices use available source or run metadata. After changing the source or its layout, refresh the preview and check selections, renamed columns and types. A failed or cancelled run may have reached only part of the flow.
Revoking source access stops future reads until you select the file again. It does not remove the original file or previously captured run data.
Runs, devices and errors
Inspect a block’s configuration and the first actionable error before rerunning. Check data shape, column types, connector compatibility, batch size and the selected device. For an unexpected comparison, keep the same source, split and seed while changing one setting at a time.
Supported devices vary by component. PyTorch capabilities may support CPU or Apple’s MPS acceleration; the scikit-learn and TensorFlow foundations use CPU. Exact saved training continuation is CPU-scoped. Unavailable combinations are refused by the app.
If a run reaches a time or memory limit, reduce the amount of data, model size or batch size, or review the allowed budget. Cancellation and failure can retain useful partial evidence; they do not mean a completed model was saved.
Exports and checkpoints
Generated Python, learned weights, trained-model packages and supported inference-program exports serve different purposes. Review the app’s export details and input/device limits before choosing a destination. An exported inference program is a model artifact with a helper, not a standalone Mac application.
Exported code is executable outside Kwip ML. Weights and packages can reveal information learned from your data. Default trained-model packages omit raw/prepared tables and fitted preprocessing state; this does not make the exported model anonymous or guarantee that it accepts a raw CSV file.
Only supported app-managed checkpoint formats and compatible architectures can be restored or continued. Kwip ML does not provide a general third-party Python or program importer.
Storage and privacy controls
Projects, source snapshots, prepared data, model values and weights can be retained on your Mac. Optional value-capture settings affect future optional observations; they do not erase earlier results or mandatory replay artifacts. Timing/memory measurements are optional.
Cache cleanup, source-access revocation, lesson-progress reset and deletion of project/run files have different effects. Check the action’s scope before confirming. Copies you exported elsewhere remain in their chosen destinations.
Read the Kwip ML privacy policy for local storage, support email and sharing details.
Third-party notices and source
Kwip ML includes open-source runtime components. Copyright, license and attribution files are included with the application. In Finder, select Kwip ML, choose Show Package Contents, then open Contents → Resources → KwipMLRuntime. Open source-disclosure → README.md for the source and license guide, or source-disclosure → SOURCE_INDEX.json for the component index. The licenses and standalone-license-evidence folders contain the notice inventory and retained license files. Python packages also retain their own license files. The applicable open-source licenses govern their covered components regardless of the application’s own license terms.
For help finding a component’s notice or corresponding source, email [email protected] with your app version and the component name. You do not need to send a project or dataset for a licensing question.
The bundled Certifi certificate list is distributed under the Mozilla Public License 2.0. Its exact source certificate list and license terms are available online. Other component-specific source pointers are retained in the accompanying notices.