The dashboard you have today
- Surfaces idle clusters, then waits for a human
- Recommendations expire while backlog priorities shift
- Savings are estimated, never proven
- Priced per seat, whether it saves you anything or not
Product
Assessment starts read-only. Autopilot stays off until your guardrails say otherwise.
The gap
FinOps recommendations age out while engineers protect roadmap work. Digital Tap turns the same signal into governed action, audit evidence, and measured savings.
How it works
Link Databricks in minutes with a scoped, read-only credential. Physically enforced: the assessment can see everything and change nothing. Your waste number appears within days.
Turn on optimization when you’re ready. Every change passes SafetyGate policy checks and carries a single-use signed token. Start in copilot (approve each fix) and graduate to autopilot.
A baseline / counterfactual engine measures every action’s real impact — with a dispute workflow. You keep 80% of every verified dollar. We earn the other 20%.
Safety
Digital Tap treats every optimization like a controlled change: scoped credentials, policy checks, signed action tokens, and shadow-mode rehearsal before autopilot.
hibernate prod-genomics-01 — after grace windowEvery mutating action is validated against your policy: protected resources, blackout windows, autonomy level, guardrail bounds.
Approved actions carry an HMAC-SHA256 token scoped to one action on one resource. Connectors refuse everything else.
Assessment mode is physically unable to write. Full visibility, zero ability to change anything — before any autonomy is granted.
The full decision loop runs against a no-op executor on your real environment before autopilot is ever enabled.
The Nerve Engine
Our agents learn each cluster’s weekly rhythm, forecast demand, and reconcile every prediction against what actually happened. Mistakes raise a cluster’s caution automatically; clean wins relax it back.
Each cluster’s utilization pattern is learned online — no batch retraining. New behavior fades in, old behavior fades out.
Hibernate into predicted idle windows. Pre-warm ahead of the Monday 9am rush. Confidence bands widen when the data is thin.
Every pre-warm prediction is written down and graded hit-or-miss against reality. Forecast skill is a number you can watch improve.
A hibernate that got resumed in minutes teaches the system to wait longer on that cluster. Recent evidence outweighs old mistakes.
Verified savings
A baseline / counterfactual engine measures every action’s real impact — with a dispute workflow. You keep 80% of every verified dollar. We earn the other 20%.
Verified savings record
DT-2026-07-19-0417
| Counterfactual — untouched | $512.50 |
|---|---|
| Actual — after hibernate | −$100.00 |
Verified savings savings_dollars | +$412.50 |
Connect read-only in minutes. Get a verified waste map in days. Turn on governed action only when the evidence is worth it.
Get the read-only assessment