Documentation ›Your AI agent
Digital employees (AI teammates)
What it is: an AI "employee" is a bot teammate that sits on your Team page right next to your people. Under the hood it's an operator of kind ai with an agent profile — its own instructions, competencies, what it may see, its authority, and its schedule.
Works across channels: you put an employee on a channel — website widget, WhatsApp, Telegram, Instagram, or phone (voice) — and it handles those conversations. One employee owns a channel's default; if none is assigned, your account's usual AI answers, exactly as before.
What its instructions are: the employee's text is its role on the line — name, character, task. On accounts that use the "About your business" block, that role sits on top of your business facts and the platform's conversation rules, so putting an employee on a line never loses your addresses, prices or hours. On the classic single-prompt setup the employee's text replaces the prompt entirely; the Team page warns you when that would shrink it.
Where: Team in the menu's Setup group — admins see it there while Digital employees is on. When it is off the menu has no link: open the Team page from this link and switch it on.
How it's enabled: the "Digital employees" master switch at the top of the Team page. New accounts have it ON; you can turn it off. Even when it's on, nothing changes on your live channels until you activate an employee and put it on a channel.
On the Team page you can:
- Hire — pick a role preset (Reception, Sales SDR, Support T1, Follow-up, Reminders, Shift Inspector, or Custom), give it a name, tick its competencies, and choose what it can see and its authority. It's created as a draft — nothing goes live yet.
- "Write the instructions for me" — describe your business in plain language (plus tone and any always / never rules) and it drafts a working system prompt for you to edit. No prompt-writing wall for non-technical owners.
- Activate / Pause — only an Active employee takes live traffic. A draft keeps the lines ticked in its hire form and starts answering on them once you activate it. Pausing takes it off every line it holds, and activating it again does not put it back — put it on its lines again.
- Put on a line / remove — assign the active employee to a whole channel or, if you have several numbers or accounts, to one specific line. A line has exactly one employee; one employee can hold any number of lines. Green chips show where it's live. The hire form has the same checkboxes, so a new employee can take free lines right away — it goes live once activated, and a line a teammate already holds is reassigned from their card.
- Active hours — "Always on" (24/7) or "Mon–Fri 9–18" in your timezone. Outside its shift, your generic AI covers.
How many you can hire. Each plan includes a number of AI employees: the trial, Office and Office + Calls one each, Business two, Pro three, Workforce as many as you need. Every employee you have hired counts — active, paused or still a draft — and a Shift Inspector counts too. When the plan's number is reached, the hire form says so.
Changing an employee after hiring. On the Team page you activate and pause it, set its hours, its lines and who it hands off to. Its name, instructions, competencies, what it can see and its authority are set when you hire it; to change them later, use the API (PATCH /api/v1/employees/{id}).
Competencies — and why they're enforced in code, not just the prompt
Competencies are what an employee is allowed to do. Each maps to real tools: answer from the knowledge base, capture leads, transfer to a human, hand off to a teammate, end the call, book an appointment, book a video demo, or draft an invoice. "What it can see" is separate: with «Nothing (knowledge only)» it answers from your knowledge and does not see what the customer talked about before; with any other choice it sees the customer's card — who they are and their earlier conversations.
The model is only shown the tools you granted, and any tool call outside that set is rejected at execution time — a jailbroken or hallucinated call can't act outside the employee's job (least-privilege).
Gotcha: a competency that depends on another feature (e.g. Book an appointment needs Calendar, Book a video demo needs Google Calendar booking) is greyed "(needs setup)" until you enable that feature first. The hire form intersects your picks with what's actually enabled, so it can never grant a tool you haven't turned on.
Authority (Advise vs Act) & hand-off
Advise (default) — the employee proposes any action that writes (capture a lead, transfer, book) and waits for your OK. Act — it performs those on its own. Draft / testing employees are sandboxed: their side-effects never fire for real.
The Advise inbox is the Approvals page: pending AI-employee actions show up there (next to co-pilot drafts) with Approve / Reject. Approve runs the action for real (race-guarded — it can't run twice); Reject does nothing.
Hand-off: with more than one AI employee, set "Hands off to" so a conversation can be routed to a teammate (e.g. Reception → Sales). The target is fixed by you, never chosen by the AI at runtime, so a hand-off can't be redirected to an arbitrary employee. The employee also needs the Hand off to a teammate competency — the Reception preset does not tick it, so tick it when you hire, or it never hands off.