FEATURE 02ROI · WASTE · POLICY
The Zombie Sentry
Stop paying for AI that isn't thinking. Reclaim GPU spend per your operator policy: alert, or — if you opt in — block in real time.
Forgotten notebooks. Runaway loops. Agents stuck firing the same kernel a thousand times a second while the bill clock runs. We watch the kernel-launch fingerprint of every PID through the eBPF ring buffer. When a process crosses your operator-set threshold, the persona decides the outcome: alert into the ledger, or — if you've armed enforcement — SIGKILL within milliseconds. Your CFO gets the GPU-hours back, and every decision is sealed in the ledger for audit.
ROI
real-time reclaim · engagement-specific
Default threshold
1000 identical launches / 2s
Latency
real-time · sub-second decision
Auto-kill
policy-driven · alert by default
// TECHNICAL SPECIFICATION
Signal
kernel-launch rate per (pid, fn_ptr, grid+block dims)
Heuristic
identical-launch sliding window · O(1) per event
Action
log alert · SIGKILL when policy = kill
Safeguards
PID-recycle + kernel-diversity guards · suppress kills on ambiguous matches
Reporting
Prometheus counters + the turn-key Grafana pane
Override
allowlist + protected_pids · tuned per-host