Welcome

A companion,
not a chatbot.

This is a walkthrough of a bedside AI companion for people living with mild-to-moderate dementia — built around non-pharmacological support for behavioural and psychological symptoms (BPSD), not conversation for its own sake.

Its distinguishing trait is that it acts before distress peaks, not just after: it watches for known warning signs and caregiver-set patterns, and steps in early.

Redirects, never corrects

Confabulation is never contradicted. The companion moves the conversation somewhere grounding instead.

Acts proactively

The register moves to elevated on a time pattern, a known precursor behaviour, or trending distress — before things peak, not after.

Stays the calm counterpoint

It never mirrors distress. When the person is upset, the companion is the steady, unhurried presence in the room.

👁

Built for caregiver insight, not just companionship — every escalation, every alert, and every field you're about to see feeds a dashboard a family member or care-home staff can actually read. This isn't a black box the person is left alone with.

What this guide covers
  1. Audio check — confirm the microphone and speaker you're using today
  2. Dashboard tour — the fields that actually shape behaviour, and why
  3. Choose a resident and companion — who you'll meet, and who meets them
  4. A live scenario — a real conversation, not a script
  5. Shape the companion — personality, a generated voice, and appearance, all optional
Before you begin

System check

A real check against this browser's actual audio devices — not a simulation.

🎧

Why a device picker — audio doesn't always route the same way through every headset. If sound stays silent on a Bluetooth or USB device, pick a different one below rather than assuming the hardware has failed.

Microphone

Try saying: "Good morning — I hope you slept well."
This checks the device only — no transcript is shown here. EVI transcribes speech once the live scenario connects.
Not tested yet

Speaker

Not tested yet
Step one

Choose who you'll meet

Three residents, each set up to demonstrate different parts of the system. The next screen tours their real profile.

Caregiver dashboard

The fields that shape behaviour

A guided tour of their real profile. Use Next in the card to step through — each stop explains what that field informs and what it influences.

Caregiver

Step two

Choose their companion

This shapes how engaging the conversation feels — and it's fully customizable later, including a voice generated from your own description.

Ready when you are

Everything's set.

Resident
+
Companion
This session is capped at 20,000 spoken characters. A live counter stays visible throughout — this is a real cap on real API usage, not a demo prop.
Live scenario — you speak, real Claude replies, real voice

COMPANIONSHIP
Usage
0 / 20,000

Suggested things to say — as the resident

You're voicing them. The companion only ever sees their real profile, never a script.

    Speaks continuously once connected — EVI listens for a natural pause, the same as the device.
    End-of-turn
    Pause override (s)
    Reminder
    Optional

    Shape the companion

    Nothing here is required. The voice tab makes a real call to Hume's voice-generation API — every preview is a genuinely new, AI-generated voice, not a preset.

    Generated by Hume Octave — not a preset

    Every swatch below already sits in the dementia-safe warm palette. The badge shows its real contrast ratio against on-screen text, computed live.

    Session complete

    That's the walkthrough.

    You saw the dashboard fields that drive behaviour, a live escalation and redirect, and how the companion can be shaped without touching a line of configuration.

    ✓ Confabulation redirect ✓ Early-warning register trigger ✓ Caregiver dashboard ✓ Real voice, generated live