AI & models · integrations
MLflow plugin — provenance for your model store
Store MLflow models, checkpoints, and datasets as verified, byte-exact AT-1 containers instead of raw files. Every artifact is compressed losslessly (never-worse than xz), carries an embedded SHA-256, and is byte-exact-verified on download — a tampered or corrupted artifact is refused, not silently returned. So an artifact in your MLflow store is provably the exact bytes you logged: the reproducibility / provenance hook regulated AI (EU AI Act, NIST AI RMF, FDA 21 CFR Part 11) needs.
Install
It's a pure-glue plugin — no engine code ships in it. It registers the at1:// artifact-repository scheme and shells out to theat1 CLI.
npm i -g @tinyfiles/cli # the at1 codec binary pip install mlflow-at1 # the plugin (registers the at1:// scheme)
Use — point an experiment at at1://
Set an experiment's artifact location to an at1:// path; everything else is normal MLflow. log_artifact, log_model, and the rest all route through AT-1.
import mlflow
exp = mlflow.create_experiment("regulated-model",
artifact_location="at1:///data/mlartifacts")
with mlflow.start_run(experiment_id=exp):
mlflow.log_artifact("model.bin") # -> stored as a verified AT-1 container
# download = a byte-exact, SHA-256-verified reconstruction:
path = mlflow.artifacts.download_artifacts(run_id=run_id, artifact_path="model.bin")What you get
- Smaller — lossless, never-worse than xz, on weights/checkpoints/datasets.
- Verified — every container embeds a SHA-256; download refuses anything that doesn't reconstruct byte-for-byte.
- Provable provenance — the artifact is the exact bytes you logged, for audit / reproducibility.
- Zero lock-in — an ordinary file with a tiny open decoder; you can always get the exact original back.
at1://<path> is a local/mounted filesystem base for the containers (S3/GCS backing is a follow-on). Reads/verify need no account; encoding is metered against the connected account — same as the rest of AT-1.