Every training run.
Every pipeline. Every cron job.
One home.
A local-first operations portal that combines a Hugging Face-style model registry with Weights & Biases-style run tracking — plus the piece neither of them offers: live monitoring for data publishing pipelines and cron jobs.
Model registry
Model cards, versions, and aliases in Hugging Face card format. Publishing goes to the real Hugging Face Hub — never a reimplementation.
Run tracking
Params, metrics, and gate verdicts for every training run, in the MLflow/W&B tradition: short names, full configs, filterable tags, honest conclusions.
Pipelines & crons
Data publishing pipelines and the full cron fleet, with freshness checks and failure streaks — the operational layer other trackers don’t cover.
Built on principles
- Local-first. Metrics never leave the machine they were computed on.
- Real data only. An honest gap beats a fabricated number, every time.
- Private by default. The operations portal requires a login; only the operator and the agent get in.