A live-demo surface for how Digitalis can listen to structured ambient cues, run the deterministic clinical safety floor, and surface the next operator question or nudge into another system through the Model Context Protocol.
Are you having trouble breathing right now?
Evidence snapshot: 2026-05-19. Demo content is simulated, contains no real patient information, and supports triage recommendations rather than diagnosis.
MCP-accessible ambient turn output
binds the request to the calling organisation
resolves existing ambient state server-side
anchors replay, audit, and cursor progression
structured reviewed cues only; no raw transcript
Are you having trouble breathing right now?
Breathing difficulty is a priority red flag and cannot be downgraded by AI reasoning.
The dummy run is an integration story: the nudge can be consumed by a call-centre screen, EHR sidebar, governance console, or orchestration agent without exposing raw transcript text.
Voice or room signals are reduced to structured consultation cues before export.
The external system calls ambient-copilot-turn with tenant, session, turn, roles, and cues.
Red flags, crisis pathways, and Australian triage guardrails constrain the response.
The UI receives the next question, disposition state, blocked reasons, and audit summary.
The investor demo still shows the older text and voice paths because they prove the same governed triage engine can route emergency, same-day care, and safe self-care presentations without allowing AI output to weaken red flags.
Use this panel if microphone, browser permission, or network access fails during the live demo.
I have crushing chest pain that started 20 minutes ago, it is going down my left arm, and I feel sweaty and sick.
Call 000 now
Red-flag cardiac safety rule sets the floor before model reasoning.
The AI can continue the handover, but it cannot downgrade this presentation.
condition records in the demo evidence snapshot
deterministic clinical controls in the validated snapshot
synthetic validation cases for regression and coverage checks
normalised evidence-reference coverage; PMID/DOI identifiers captured for 98.4% of tuples
No raw patient transcript leaves the governed ambient context.
The realtime voice path remains available for a live walkthrough.
Provider keys are never exposed to the browser.
Readiness, smoke output, and assurance gates remain linked from the demo.
Deterministic rules execute first and later AI reasoning cannot override red flags.
Clinical recommendations, evidence, and handover artefacts are reviewable by governance teams.
Deployment status, smoke tests, and release gates are linked from the demo path.
The platform now demonstrates ambient MCP access, safe escalation, right-place care navigation, voice and text intake, auditability, and operational visibility.