The two assignable roles
The industry calls both humans and bots “agents”. In these docs, agent always means a person;
the software is always an AI agent. The
agent role above is the human one.A third role,
manager, exists in the data model but is not assignable from the dashboard. You may
encounter it through the API. It is not currently part of the product’s permission model — if you need
something between agent and administrator, that is what custom roles are for.Custom roles
When the built-in roles don’t fit — a finance reviewer who should see reports but never touch conversations, an outsourced team that should only handle unassigned tickets — you can define a custom role from six permissions.
The three conversation permissions are deliberately graded rather than a single on/off.
conversation_participating_manage is the tightest — useful for contractors or overflow teams who should see their own work and nothing else.
Custom roles are an Enterprise capability. See Custom roles for setup, and
Plans for what each tier includes.
What roles do not control
Two things people expect to find here and won’t: Tool scope. What your AI agent can touch in your connected systems is set by your plan and your integration configuration, not by human roles. A custom role cannot grant an AI agent write access, and an administrator’s permissions do not widen what the agent may do. See Plans. Billing visibility. Access to billing and invoices follows theadministrator role specifically; it is not one of the six custom-role permissions.
Changing someone’s role
Roles are assigned when you invite someone and can be changed at any time from workspace settings. Removing a person’s access does not delete their conversation history — past assignments and replies stay attributed. See Users and roles.Next
Users and roles
Inviting people, assigning roles and deactivating access.
Teams and business hours
Grouping agents for routing and coverage.