”Give me a complete audit of everything running in my workspace.”
The agent inventories all resources:list_instances— all GPU instanceslist_volumes— all storage volumeslist_environments— all environmentsbilling_balance— current credit balancebilling_usage— spending over the last 30 days
Infrastructure Audit — May 19, 2026 Instances (4 running):Volumes (3 total):
Instance Environment GPU Uptime Cost/hr prod-inference production H100 12 days $2.50 staging-test staging A100 3 days $1.60 dev-box main RTX 4090 18 hrs $0.35 old-benchmark main RTX 4090 9 days $0.35 Environments: main, staging, production Balance: 1,892.40 Current burn rate: 115.20/day)
Volume Environment Size Region Status model-weights production 500 GB us-east-1 attached training-data main 200 GB us-east-1 available temp-scratch main 100 GB eu-central-1 available
“What’s wasting money? Where can I cut costs?”
The agent analyzes utilization:ssh_executeon each instance — checks GPU utilizationlist_volumes— checks for unattached volumes
Waste identified:Potential savings: $213.30/month (45% of current spend) Kill old-benchmark first (idle 9 days), then shut down staging-test between sessions, and delete temp-scratch (wrong region).
Resource Issue Monthly cost Action old-benchmark 0% GPU utilization for 9 days $75.60 Terminate staging-test 2% GPU utilization $115.20/mo Terminate when not testing temp-scratch Unattached volume, different region $7.50/mo Delete if not needed training-data Unattached volume $15.00/mo Delete or attach
”Clean up the waste. Kill idle instances and delete unused volumes.”
The agent executes the cleanup:delete_instance— terminatesold-benchmarkdelete_volume— deletestemp-scratchbilling_balance— checks updated balance after refunds
Cleanup complete:Updated burn rate: 4.80/hr) Monthly savings: ~$252 Remaining resources: 3 instances, 2 volumes across 3 environments. Next recommendation: shut down staging-test when you’re not actively testing.
- Terminated old-benchmark (idle 9 days) — saving $0.35/hr
- Deleted temp-scratch (100 GB, eu-central-1) — unused credits refunded
Tools used in this workflow
| Tool | Purpose |
|---|---|
list_instances / list_volumes / list_environments | Full resource inventory |
billing_balance / billing_usage | Credit and spending data |
ssh_execute | Check GPU utilization per instance |
delete_instance / delete_volume | Clean up waste |