For ML engineers
Can you prove what your
agent just did?
Run this once. See everything that actually happened.
$uv pip install roar-cli
Run this on your last training job.
then initialize
$roar init
then run it
$roar run python train.py
Captures every input, output, dependency, and environment — automatically.
This is what gets captured
[ DAG image — set src on #dag-img ]
Every run becomes a complete, queryable lineage graph.
Your agent just ran something.
Can you answer:
- what data it used?
- what changed since the last run?
- can you reproduce it?
If not, you don't have a result.
You have a one-off.
No changes to your workflow
No code changes
Works with PyTorch, JAX, TensorFlow
No YAML
Works with Ray / multi-node
No account required
<3% overhead
Ready to see it?
$uv pip install roar-cli
Run this on your last training job.
Where did you hear about roar?
Thanks.