Beta v0.1
Stop the GPU Bleed.
Read-only Observability for ML Infrastructure.
Find idle GPUs, stalled training runs, and cloud waste automatically. Zero write permissions. Zero infra risk.
Get Early Access
Read-only. Nothing changes without your OK.
Trace Diagnostics v0.1.0
$trace scan --cloud runpod --region us-east-1
[INFO] Authenticated via Read-Only API Key...
[INFO] Scanning 12 active A100-80GB instances...
[WARN] Idle GPU Detected: instance-7f2a
- Utilization: 0.2% (last 4 hours)
- Process: none detected
- Estimated Waste: $3.80/hr
- Process: none detected
- Estimated Waste: $3.80/hr
[WARN] Stalled Job Detected: training-run-beta
- Error: CUDA Out of Memory (OOM) at Epoch 14
- Status: Container running but idle
- Status: Container running but idle
$_
Built for the ML Stack
Deep integration with your high-performance compute.
Idle Detection
Automatically flag GPUs running at <1% utilization for more than 30 minutes. Stop paying for forgotten instances.
Zero-Write Access
We only request read permissions for pod status and billing metrics. No SSH access. No execution rights.
Root Cause Trace
Link CUDA crashes directly to the GitHub commit that caused them. Debug failures in seconds, not hours.
Ready to Trace your cluster?
Join the first 1,000 ML engineers getting deep visibility into their cloud spend and cluster health.
Get Early Access