AI Encounters & Kiosk
An AI encounter is a structured conversation between a patient and an AI assistant, set up by the service. There are three kinds:
| Kind | What it is for |
|---|---|
| History taking | Collecting the patient’s history before they see a clinician. |
| Health education | Explaining a condition, procedure or preparation from the service’s content. |
| Financial counselling | Walking through costs, schemes and payment options. |
A patient completes an encounter either remotely on the patient portal, from a link the service sends, or in the clinic on a device a staff member hands them (the kiosk), by text or by voice.
Settings
On the service’s Settings tab, the AI encounters card sets up each kind the service offers. For each kind:
- Enabled: whether the service offers this kind.
- Instructions: anything specific to add to this encounter’s conversation.
- References & protocols: documents from the content library attached by default. Only processed documents can be chosen.
- Share the care team’s latest note with the assistant: folded into the assistant’s context when the patient opens the encounter.
- Auto-attach relevant library documents: searches the library when the encounter starts and attaches the best matches (up to three) alongside the defaults.
- Let the assistant search the service library during the conversation.
Encounters no longer need an agent bound to them: the assistant is set up from these settings, the service context and the patient’s own context.
Sending an encounter to a patient
On the patient’s page, General tab, the AI encounters card:
- Click Start encounter.
- In Start an AI encounter, choose the Kind, optionally say Why this encounter (it “gives the assistant a head start instead of asking the patient to explain from scratch”), and adjust the References & protocols (pre-ticked from Settings).
- Click Send link.
The link goes out on the service’s default channel (see Channel). If sending fails, Encounter started — link not sent shows the link with a Copy button so you can pass it on yourself.
The patient can also be sent an encounter by the patient agent with the Hand over for a full encounter tool.
Following an encounter
The card lists every encounter with its status (Pending, Active, Completed, Abandoned, Cancelled or Expired), who started it, whether it was Shared by patient, and its summary. On each row:
- the transcript icon opens the full conversation
- Cancel stops an encounter that has not finished
- Resume on this device reopens an in-clinic encounter on the current device
What the assistant was told
The transcript dialog has a What the assistant was told panel, so staff can check exactly what the assistant knew:
- when the context snapshot was built
- each reference, badged Chosen (picked when starting), Default (from Settings) or Auto (auto-attached)
- the Service context
- the Care team plan for this patient
The care plan context
When the patient opens an encounter, Medbot builds a short care-plan briefing for the assistant from the service and the patient’s record:
- why the care team started this
- where the patient is in the service
- the care team’s briefing
- the latest note from the care team (if enabled in Settings)
- upcoming appointments
- open care-team actions
- who is on the care team
It is kept to about 5,000 characters so the assistant stays focused.
In the clinic: the kiosk
For a patient who is in the clinic, a staff member can run the encounter on a clinic device such as a tablet.
- On the patient’s AI encounters card, click Start here on this device.
- In Start an in-clinic encounter, choose:
- Kind
- Conversation mode: Text or Voice
- Assistant look (voice only): Let the patient choose, Orb or Face
- Why this encounter (optional), and the References & protocols
- Click Hand over the device.
The device switches to a full-screen kiosk (/kiosk/...) with no staff app around it.
- The patient is first asked for consent: Yes, start or Not now.
- In text mode they type; in voice mode they talk, with captions showing Listening…, Thinking…, Speaking… and “Ready — go ahead and speak.” In voice mode the patient can switch between the Orb and the animated Face, and can fall back to text.
- End session finishes the encounter.
- Staff: exit returns the device to the staff app (“Leave this encounter?”). The patient’s progress is saved, and the encounter can be resumed on this or another device from the patient’s page.
Requirements for voice
Voice mode needs a default realtime model (set by an administrator under Realtime Models, /admin/realtime-models) and the AI Gateway to be configured. If either is missing, the dialog says why voice is unavailable and text still works.
Channel settings
On Settings, the Channel card sets how this service reaches patients:
- Default channel: WhatsApp or SMS, for AI-encounter invites and appointment reminders. SMS is used automatically if WhatsApp is not configured or a send fails.
- Send appointment reminders and Hours before the appointment (e.g.
24). See Appointments.