Train with any Python machine learning library, read training data from the warehouses and buckets where it already lives, deploy prediction APIs where your engineers already run services, and get alerts where people already look.
Fit or train a model with any of these and maclnote records its hyperparameters, per-epoch or per-fold metrics, and saved weights with no code changes. Any other Python code is recorded through a two-line explicit call.
Connect once per source with your existing credentials. When a notebook reads a table or file for training, maclnote records a fingerprint of exactly those rows, so a large dataset is versioned without copying it and every model knows what it was trained on.
By default a deployed model answers prediction requests from maclnote's cloud. Team and Enterprise customers can instead deploy into their own Kubernetes cluster or cloud account, with the same one-click deployment and monitoring.
Notebooks sync to Git as clean diffs. Drift alerts, model approval requests and training run summaries go to the channels people watch. Log in with your company identity provider, and everything in the interface is also available through the API.
Import your existing training runs, hyperparameters, metrics and saved models with one command so your experiment history comes with you. Notebooks that already call the MLflow logging API keep working unchanged; maclnote accepts those calls alongside its automatic recording.
Tell us which framework or data source you use. Anything with a Python client works today through the SDK, and native connectors are added in order of requests.