AI-powered auditing for rented GPUs
You pay for every node, every month. Afterdrain counts the ones you got.
The hours you paid for and did not get, proven from your own logs, filed before the credit window closes.
Get a free capacity statementThe provider holds the only ruler.
We think every renter should hold their own.
- Billed: 8 of 8
- Provider counts: 8 of 8
- You count: 7 of 8
1 paid for, not delivered
Illustrative.
What you get, every month.
statement · this month
hgx-a7-0412
Xid 79 · fell off the bus · 41h to replace
Slurm event · Xid log · ticket #4471
Counts against provider
Hard down, no maintenance notice
hgx-b2-0188
Drained by our own health check · 6h
Slurm event · NHC reason · our ticket
Excluded
Renter-initiated drain
hgx-a3-0091
Up, but NVLink degraded · 9 days
DCGM NVLink counters · NCCL errors
Counts against provider
Healthy-node clause in the order form
Built for teams renting 64 to 2,000 GPUs.
- 01.AI labs training on reserved clusters.
- 02.Robotics and physical AI teams.
- 03.Bio and science teams running big jobs.
On Slurm or Kubernetes, from any GPU cloud.
What did you actually get last month?
Send one export. One statement back in a week. Free.
Get a free capacity statement