The product

Four moves, repeated every second.

PACE is a loop: meter what is happening, predict where it goes, intervene before it gets there, then explain it in language a human can act on.

I

Meter

Every tool call, token, spawn and retry is metered against the task that requested it — not the resource that served it. Attribution is the whole game.

  • per-agent burn rate
  • cost-per-task ledger
  • classic resource sweep
II

Predict

Prophet-based seasonal models separate steady infra from spiky agent workloads, so a 6σ token burst doesn't hide inside a normal-looking cloud bill.

  • 7 / 30-day forecast
  • anomaly z-scoring
  • what-if simulator
III

Intervene

The part no other tool does: PACE stops things. Hard caps, spawn limits and a breaker that pauses a process mid-flight, before the invoice becomes history.

  • circuit breaker
  • spawn limiter
  • approval-gated playbooks
IV

Explain

A small local model narrates what happened in plain English and hands you the rightsized Terraform. No API key, no token bill inside the cost tool.

  • spike explainer
  • Terraform generation
  • governance reports

The demo is live and the data is honest about being fake.