AI in Heltio writes drafts of documentation: suggested history questions, a draft visit summary, an outline treatment plan. Every result is labelled AI-generated draft — review before accepting, and nothing is saved without your click. The clinical decision stays with you.
What works today:
- Anamnesis copilot — suggested questions during a visit. See Anamnesis copilot.
- Appointment summary draft — after the note is signed. See Appointment summary draft.
- Treatment plan suggestions — an outline of phases and goals from the diagnosis text. See Treatment plan suggestions.
- AI Scribe — a recording of the visit turned into a draft SOAP note. The panel opens in the active session and needs the patient's separate consent to being recorded.
- AI briefing — a short pre-visit summary, available from the appointment list 4 hours before the slot.
- Handwritten notes OCR — does not work yet; the toggle exists in settings, the scanning screen does not.
How to do it
Open Settings → AI Features
The AI Features page shows three things: the Enable AI features toggle, the list of individual features, and the credit balance.
The toggles are read-only. Heltio enables AI on the server side — hovering shows "AI activation is managed by Heltio. Contact support to get started." Write to support if you want AI running in your clinic.
Accept the data-processing agreement
Before AI can run, someone with admin rights has to read and accept the AI Sub-processors — Data Processing Agreement. Click View disclosure, scroll to the end (until then it reads "Please scroll to read the full disclosure.") and confirm I accept and want to enable AI features for this clinic.
Once saved, the date stays on the page: Data Processing Agreement accepted on.
Check the credit balance
Every AI call spends clinic credits — from 1 for a batch of questions to 8 for a treatment plan. You see the cost on the button before you click. Details in AI credits.
Review the activity
The AI Usage tile leads to the AI Activity page — a list of every call: feature, practitioner, patient, credits, time and status. You can filter by feature, practitioner, outcome and date range.
What AI in Heltio does not do
- It does not diagnose and does not recommend medication. Model responses pass through a safety filter; anything that crosses that line is discarded and the credits come back.
- It never sends anything to a patient on its own. Every message needs your click.
- It does not see personal data. Before anything goes to the provider, names, national ID numbers, addresses, phone numbers and the clinic name are replaced with tokens, and the response is translated back on our side.
- It does not learn from clinic data. The provider agreements specify zero retention and no model training.
- One exception: audio. A visit recording cannot be de-identified before transcription — it reaches the provider unchanged. That is why recording needs the patient's separate, documented consent.
When something doesn't work
AI features are not enabled for this clinic.
What this means: The clinic does not have AI switched on yet, or the processing agreement has not been accepted.
AI features are temporarily disabled system-wide.
What this means: Heltio paused AI for every clinic — usually while a provider is being worked on.
This AI feature is currently disabled. Enable it in AI Settings.
What this means: AI runs in the clinic, but this one feature is switched off.
AI provider is temporarily unavailable. No credits were charged.
What this means: The model did not answer in time, or returned an error.
The AI output did not pass safety review and was discarded. No credits were charged.
What this means: The model started writing something the assistant is not allowed to write — a disease name, a medication, or an interpretation of a test result.
AI rate limit reached. Please wait a moment.
What this means: Too many calls from one account in a short time.
AI features require an internet connection.
What this means: The app is running offline.
Who has access
Clinic Admin PractitionerThe AI Features page, the credit history and AI Activity are visible to the owner and the clinic admin — other roles see "Only clinic owners and admins can manage AI settings." The assistants themselves are run by the practitioner conducting the visit.
Related
- AI credits — balance, costs and history.
- Patient consent for AI — what the patient must know and what they agree to.
- Audit log — the trace left by every AI call.
Related features
AI credits: balance, costs and history
Every AI feature draws on one clinic balance. Here is what each call costs, when credits come back, and where to see the history.
Anamnesis copilot
A side panel in the active session suggests the next history-taking questions based on what is already in the note. You decide which one to ask.
Appointment summary draft
Once the clinical note is signed, AI fills in a draft summary: exercise recommendations from the clinic library, lifestyle advice, a follow-up plan and a safety checklist.
Treatment plan suggestions
From a diagnosis description, AI lays out an outline therapy plan — phases, progression criteria and measurable goals. You review and correct everything before the plan exists.
Consent for AI processing
Two consents, not one: the clinic accepts the data-processing agreement, and the patient — separately, and only for recording — agrees to the visit being recorded.
Handwritten notes OCR
Reading text from a photo of a note is planned but does not work yet. Today you attach the photo to the appointment and type the content in by hand.