Skip to main content
Ask the assistant to make a change, not just explain it, and it will offer to do it — “only send chats to agents who are Ready”, “create a morning shift 9 to 5, Sunday to Thursday”, “give Sales Team B the support channel”.

​
You never need to know an id

Refer to things the way you normally would — “the Sales role”, “the morning shift”, “Ahmed”. The assistant looks them up itself and finds the right one. If more than one thing matches, it asks which you meant by name. If nothing matches, it says so and lists what does exist. It will not ask you for an id. Looking things up is read-only and needs the matching view permission, so it can only see what you could already browse in the app.

​
Nothing happens without you

The assistant cannot change anything on its own. When it proposes something you get a card showing exactly what would change, with the current value beside the new one:
Those values are read from your workspace at the moment it proposes, not from what the assistant says it will do. If its explanation is wrong, the card still shows the truth. Nothing is applied until you press Apply. Dismiss discards it.

​
It can only do what you can

Every action requires the same permission as doing it by hand: If you do not hold the permission, the assistant is never offered the action at all — it will explain how to do the thing instead, and you can ask an admin. Your permission is checked again when you press Apply, so an approval cannot outlive your access.

​
Asking versus telling

The assistant is deliberately conservative about this:
  • “How do I only send chats to Ready agents?” — a question. You get an explanation.
  • “Only send chats to Ready agents.” — a request. You get a card to approve.
If it is unsure, it answers the question and offers to make the change rather than proposing one you did not ask for.

​
Diagnosing a problem

You do not have to describe your setup. The assistant can read it. Ask “why isn’t Ahmed getting any chats?” and rather than handing you a checklist, it looks up his roles, their data scope, the channels each role grants, and the workspace distribution settings — then tells you which of those is the cause. It checks, in one pass: It reports every cause it finds, not just the first — these stack, and fixing one while two remain is why “we already checked that” happens.
If the configuration is clean it says so, and points at the two things it cannot see: current presence, and open-chat load. Distribution gives new conversations to whoever has the fewest open chats, so an agent whose colleagues are lighter will legitimately be skipped. See How assignment works.

​
Why sending stopped

The other half of the support load, and the one no amount of channel troubleshooting surfaces: plan limits fail silently. The unit pool runs out, outbound messages stop, and incoming messages keep arriving — so the inbox looks perfectly healthy while nothing you send leaves. Ask “why aren’t my messages going out?” and it checks: Bonus units from a growth package count toward the allowance, and an unlimited plan is never reported as exhausted. If nothing at the plan level is wrong, it says so and points at the next likely cause: the channel’s connection status, and whether the 24-hour customer service window has closed — outside it only an approved template can be sent.

​
Reading the facts directly

When you want data rather than a verdict, it can also read back:
  • A whole account — email, status, roles, each role’s data scope, and exact channel access
  • Everyone and their roles, optionally narrowed by a search
  • The workspace inbox settings — the distribution and presence configuration
  • The plan — limits, which features are enabled, usage this period and SMS credits
All of this is read-only and permission-checked: it can only see what you could already open in the app yourself. Nothing here changes or deletes anything — the assistant cannot alter a setting without showing you a confirmation card first.

​
What it can do

The five distribution toggles, and the presence thresholds — idle, auto sign-out and the offline grace period.
Create a shift, change its hours, days, grace minutes or default flag, replace its roster, or delete it.
Create a role, duplicate one (the documented way to build a team), set which channels a role covers and whether that includes groups, and give a user a role.
Invite someone by email with the role they get on acceptance — if that role has no channel access the card says so, since that is the most common reason a new agent receives nothing. It can also suspend or reactivate an account.
Create an agent with its instructions, set the default, and turn one on or off. Assigning an agent to channels, uploading knowledge and training still happen in the agent wizard.

​
Deleting things

Destructive actions look different — they are styled as a warning, spell out the consequence (“3 agents on this shift will fall back to the default”), and require you to type a word rather than just click.
Deleting a role or a user is not something the assistant will do at all — suspending is offered instead, which is reversible. Removing a role strips access from everyone holding it, and removing a user needs their conversations and customers reassigned first — steps that belong on the real screens, where you can see what you are affecting.

​
Everything is logged

Every applied change is written to the audit log, attributed to you, with the mechanism recorded — so it reads “changed via the assistant” rather than hiding how it happened. Failed and refused attempts are logged too.

​
If applying fails

Somebody else changed the same thing while the card was on your screen. Applying anyway would overwrite their change with a value you never saw, so it is refused. Ask again to see the current values.
Your access changed between the suggestion and the approval. Try Recheck access in the user menu.
Suggestions last ten minutes and can only be applied once. Ask again.