Skip to main content
Every template lands on the canvas fully wired — trigger, steps, safety playbooks and, for many, a starter test suite — and you edit from there. Nothing runs until you deploy it, and anything that sends or charges pauses for human approval unless you deliberately remove that step. A template’s trigger tells you how it lives: Chat agents answer when someone talks to them, Scheduled ones run on their own clock, Event ones fire the moment something happens in your workspace, Webhook ones accept signed calls from outside, and Voice answers the phone.

​
How templates tell your team

When a template needs a person, it does two things, and neither one names the patient:
  • An alert worded from a fixed reason — for example Billing question needs a person or Clinical question for the care team. It may add a short note with a count, such as “3 claims need review”, but never a name, contact detail, date, record or claim number, or health detail, because the whole team sees it. See Notices from your AI agents.
  • A task that says what needs doing. When the work is about a patient, the task is linked to that patient’s record and appears on the calendar for its due date — open the task, then the linked patient, for the details. When it is about a lead, it lands in the CRM against that contact. When it is about nobody in particular — a slot that opened, a stock alert — it is a task for the team.
If somebody who is not yet a patient texts, calls or writes in, the task has no link; the conversation itself is on the agent’s Runs page. A task or note that would identify someone is refused and the agent is asked to try again without it. The check catches what the agent already knows about the person and anything shaped like a contact detail, date or number; it cannot recognise every way of describing someone. Agents you installed before this change keep the instructions they were installed with. To pick up the new wording, install the template again, or edit the agent’s steps to match.

​
Billing & payments

The set most practices deploy first: money follow-up that happens every day whether anyone remembers or not. How to use them well: deploy the digest first — it changes nothing and shows you the money sitting still. Add the two recovery agents next; they only ever react to failures. Bring in the assistant last, once your team is used to approving its drafts — it is the only one of the four that can move money, and it always asks first.

​
US insurance & claims

These spend insurance credits for the checks and claims they run, exactly as if a person had run them.

​
Front office & assistants

​
CRM & growth

​
Documents, intake & scheduling

​
Clinical & operations safety nets

​
Mail

How to use it well: create the labels you actually file by first — it uses yours and will not invent one. Give it a week of reading its drafts before you trust them; a draft it wrote is marked AI-assisted in the conversation, so you can always tell. See Read your mail.

​
The public page, from the visitor’s side

When a template like the Front desk assistant is published, the page a visitor lands on is a complete product in itself:
  • Your branding — the title, headline, welcome message, accent colour, theme and logo you set, on your own link (or your own domain).
  • Suggested prompts — up to six one-tap questions, so a visitor who does not know what to type still gets somewhere.
  • Real answers, safe boundaries — the agent answers from your clinic’s public information and can hand the visitor to booking or the waitlist; it holds no patient data, and a symptom question gets a polite deferral to booking, never advice.
  • An email gate, if you want one — Require email to chat identifies visitors before their first message, and disposable addresses are refused.
  • Your caps, always — per-minute and daily per-visitor limits and a monthly run cap keep an open link from becoming an open wallet.