Private Beta · Free Access

Never lose a
training run again.

mlvault snapshots your ML artifacts (checkpoints, logs, metrics) encrypts them before they leave your machine, and stores them permanently alongside your MLflow runs.

Apply for beta access See how it works

One pip install  ·  Works with existing MLflow setup  ·  Free during beta

You've been there. Three days of training. The instance goes down. The checkpoint is gone.
How it works

Set up in five minutes.

Works alongside MLflow. No migration. No new workflow to learn.

1

Install and initialise

One pip install pulls everything in. Paste your storage credentials and you're ready.

pip install obsideo-cloud-mlflow
mlvault init
# Paste your mlvault storage credential when prompted.
# Get one by applying for beta access below.
2

Train normally with MLflow

No changes to your training loop. Log metrics and artifacts the same way you always have.

import mlflow

mlflow.set_experiment("my-experiment")

with mlflow.start_run() as run:
    mlflow.log_metric("val_loss", 0.42)
    mlflow.log_artifact("checkpoints/epoch_10.pt")

    run_id = run.info.run_id
3

Commit your artifacts

Encrypts on your machine, then stores permanently. Takes 30–60 seconds. Run it in the background while your next experiment starts.

mlvault commit <run_id>
# Encrypting...  ✓
# Uploading...   ✓
# Done. Artifacts committed: my-experiment_20260307_143022_a3f9c1
4

Restore anywhere, any time

Download, decrypt, and reconstruct your full artifact directory from any machine.

mlvault restore my-experiment_20260307_143022_a3f9c1
# Restored to ./my-experiment_20260307_143022_a3f9c1/
Why mlvault

Built for how ML teams actually work.

Not generic file storage bolted onto a tracker. Purpose-built for training runs.

🔒

Encrypted by default

Your artifacts are encrypted in transit and at rest. Enterprise-grade encryption applied automatically, no configuration, no S3 bucket policies, no IAM roles. It just works.

♾️

Durable storage

Artifacts aren't tied to your cloud account, your instance, or a single provider. Stored across distributed infrastructure: resilient, redundant, accessible when you need them.

MLflow-native

Keep using MLflow for experiment tracking, metrics, and the UI you already know. mlvault adds permanent encrypted artifact storage as a plugin. Nothing to migrate.

💰

Cheaper than S3

Flat-rate pricing. No egress fees. No surprise bills at the end of the month. The more artifacts you store, the bigger the saving over AWS.

Provider Cost / TB / year Client-side encryption MLflow native
mlvault ~$150 ✓ Default ✓ Plugin
AWS S3 ~$276 ✗ Server-side only ✗ Manual setup
Google Cloud Storage ~$240 ✗ Server-side only ✗ Manual setup
W&B Artifacts $600+ (per user) ✗ Separate platform
Who it's for

Made for people who train models seriously.

ML engineers

Running experiments on GPU workstations or cloud VMs who need checkpoints to survive instance termination.

Research teams

Who need to reproduce experiments months later and can't afford to lose the artifact trail.

AI startups

Training proprietary models on sensitive data who need encrypted storage they actually control.

Free access. Limited spots.

We manually provision every account, so you get real support, not a ticket queue. That's the beta deal.

We review applications within 48 hours.