UCP
Universal Control Plane
Gates every AI action — admits or refuses it before it fires, with a signed receipt either way.
Closes F4 · Prevent pilot
See Universal Control Plane →
Mathematically Governed AI
AI you can prove.
AI you can prove.
Not AI you have to trust.
We build AI that proves every decision before it acts — the same answer every time, signed, and verifiable by your auditor without us in the room.
PROOF
SMARTHAUS · DECISION RECEIPT
Was this action allowed?Yes
Which invariant applied?Refund policy v9
Same answer as last time?Yes · byte-identical
Replayable by your auditor?Yes · clean machine
Has the model been swapped?No · hash matches
When was it decided?2026-04-30 14:22 UTC
Signed?Customer-held key
Signed by a key SMARTHAUS does not hold. The receipt travels with the response. Your regulator can verify it without contacting us.
Built for the controls the regulator is asking for · As of 8 June 2026
Where this is going
The guess is leaving the screen. Now it can cost lives.
AI is moving onto factory floors, into vehicles, onto surgical tables. In a 2025 factory test, a Unitree H1 humanoid suffered a control-software failure and lashed out at the engineers beside it. No one was hurt — the tether is the only reason it is a clip and not an obituary. The same guess that costs a refund today drives a motor tomorrow.
Six failures. One cause.
AI models are probabilistic. They guess. Every failure that kills a pilot traces back to that one fact. The industry's answer is to bolt another AI on top to watch the first one. You cannot govern a guess with another guess. So we inverted it: we prove the math before we write the code.
The discipline that makes that possible is the Mathematical Autopsy. See how we build →
We stop the action before it fires.
Wrapper-style governance · the rest of the market
The control runs after the action.
- A probabilistic classifier inspects the output once it has been produced.
- The dashboard alerts your team after the trade, the denial, the email has fired.
- Evidence is a log of what happened. Reconstruction depends on the vendor.
- The control itself is AI — it has its own error rate and its own failure modes.
- By the time their tool sees a bad output, the action has already committed.
SMARTHAUS · mathematically governed runtime
The control runs before the action.
- We admit the action — we decide, before it fires, whether to let it happen.
- A gate checks the action first, and gives the same answer every time for the same input (it is deterministic). It checks the rules the action must never break — its invariants — proven before launch in Lean 4, a public proof system.
- If a rule would break, the action never fires. That is built into the code, not checked after the fact (by construction) — and never left to another AI's judgment.
- The evidence is a signed receipt for each decision, replayable on a clean machine without us.
- The gate is mechanical. It cannot be talked out of it by a confident, wrong answer.
Who checks it
Every decision has to hold up in front of someone who can pull the thread.
There are five of them. One proof object — the Decision Receipt — answers all five, without us in the room.
The auditor
“Show me what this decision checked — and prove it ran.” The receipt names the rule, the inputs, and the result, re-runnable on a clean machine.
The regulator
“Prove the control applied before the action.” The signed receipt is evidence the gate fired first, not a log written after.
The board
“Can we defend this if it goes wrong?” One bound record per decision — not a vendor’s assurance.
The customer
“Was I treated the way the policy says?” The same input returns the same governed answer, every time.
The court
“Produce the record.” Signed with a key we do not hold — verifiable without ever contacting us.
What we sell
Four products. One layer ties them together.
What we sell is four products — a control plane and an inference engine you run, a custom runtime we build for your hardest capability, and the engine you build it all with. All inside your environment: your compute, your keys. Above them, Operation Center turns every receipt into evidence a regulator will accept.
SAID
Sealed AI Deterministic Inference
Same input, same mathematically-governed output — every call replayable and signed, verifiable by your auditor.
Closes F2 · Replay pilot
See Sealed AI Deterministic Inference →
MAE
Mathematical Autopsy Engine
The engine you build proven software with — run the autopsy method in your own pipeline.
The build engine · powers MGR
See Mathematical Autopsy Engine →
MGR
Mathematically Governed Runtime
Your hardest capability as a governed, deterministic runtime — proven in math during the engagement, then yours to keep and run.
Closes F1·F3·F5 · Specify pilot
See Mathematically Governed Runtime →
You cannot govern a guess with another guess.
Build it before the failure that forces it.
The clock is still running — Colorado SB 26-189 · Jan 2027 · EU AI Act high-risk · 2027 · and SR 26-2 just handed AI governance back to you.