One engine. Any regulated workflow. From pilot to production-trusted.
The pilot's model reads the file and drafts a decision. The GateFrame engine then applies deterministic, inspectable controls — policy thresholds, screening dispositions, completeness, confidence gates — and routes what it can't clear to a human. The governance layer is not itself a black box, and the institution keeps sign-off. Pick a workflow and run it — the same engine, reconfigured.
Pilot · probabilistic
An LLM reads the documents and drafts the decision.
Engine · deterministic
Auditable controls govern that decision against policy.
Output
A production-readiness record — cleared, or routed to a human.
Workflow:↳ Same engine, same controls framework — only the policy config and the case change. That is the reuse.
Live mode calls Claude to draft the pilot decision on this file from your browser. For local demos with your own key — don't ship a key on a public site.
Pilot
→
Harden
→
Assured
→
Production-ready
The case —
01 · Pilot AI — the decision the pilot produces today
Reading the file…
Pilot model · unaudited output
02 · Assurance engine — deterministic controls run against the configured policy
03 · Production-readiness record — what the reviewer receives
Ready for human reviewNot auto-cleared
Held for sign-off
Final decision: GateFrame delivered the file production-ready, every item evidenced and traceable.Engine configured once · reused on every case
Audit record · ordered, reviewable, hash-linked
North-star metric · AI's share of the work
Human-review time falls as the engine clears the routine cases.
Auto-cleared — full audit evidence, no human touchRouted to human — exceptions like this one