What it does This repo houses the legacy Python client for Neptune, an experiment-tracking platform that logs metrics, hyperparameters, and artifacts from ML training runs. It integrates with the usual suspects—PyTorch, TensorFlow, Keras, XGBoost, LightGBM, Optuna, and RL frameworks—so teams can compare runs and reproduce results later.

The interesting bit The README is essentially a forwarding address. A big “IMPORTANT” banner points users to neptune-client-scale, the new client for “Neptune app version 2.x.” The old 2.x docs are now “legacy.” That’s a rare public admission that the original architecture couldn’t keep up with the workloads it now markets to.

Key highlights

• 622 stars, 20 tagged topics spanning the full ML stack
• Integrations with most major frameworks (PyTorch, TensorFlow, Keras, XGBoost, LightGBM, Optuna)
• Explicitly labeled as legacy; active development has moved to neptune-client-scale
• Legacy documentation lives at a separate subdomain (docs-legacy.neptune.ai)
• Support still available through dedicated support center

Caveats

• The README reveals almost nothing about current functionality—it’s a landing page, not documentation
• No version numbers, changelog, or migration guide in the repo itself
• If you need the new “foundation model training” features advertised in the GitHub description, you are looking at the wrong repository

Verdict Worth a bookmark only if you’re maintaining an existing Neptune 2.x setup and need the legacy docs. Everyone else should follow the banner to neptune-client-scale.