Human support & handoff

Some conversations need a person. AI Assistant lets a user reach a human, and lets a member of your team step into the same conversation — no re-explaining, no new channel. The assistant and your team share one thread.

How a user asks for a human

Your user simply says so. When they ask to talk to a person — "can I speak to someone", "get me a human", "I need an agent" — the assistant recognizes the request and raises a handoff, rather than trying to muddle through. The user stays right where they are; nothing hands them off to a separate widget or email.

How your team joins

On your side, an operator inbox collects conversations waiting for a human. A team member opens one, sees the full history the user and the assistant already built, and takes over — replying inline in the same thread. To the user it's seamless: the same conversation, now with a person answering.

When your operator is done, the conversation can return to the assistant — the handoff is a moment in the thread, not a dead end.

Assignment and operator notifications

Choose manual claim, round robin, or least active in the workspace Console. Automatic modes consider only active workspace members who are currently marked available and still have capacity. Every decision records the algorithm, eligible-operator count, selected operator, and measured load so an unexplained or fake assignment cannot appear successful.

The durable in-app inbox is the primary operator channel. For an automatically assigned request, the notification is addressed only to that operator and uses the existing notification fanout for linked Web Push, APNs, and personal Telegram endpoints. A provider leg becomes delivered only after at least one endpoint reports a successful send; an in-app notification with no opted-in endpoint remains accepted.

Email, a tenant-authored Telegram destination, and webhook routes can also be configured. These are explicit queued extension points: they stay queued until your deployment provides a credentialed delivery worker, and AI Assistant never labels them delivered merely because they were configured. The delivery history keeps queued, dispatching, accepted, delivered, failed, and skipped distinct.

Turning it on

Human handoff is opt-in. It's off until you enable it for your tenant, because it depends on your team being ready to answer. When you turn it on (in Governance & model policy), the assistant starts offering the path to a human and your operator inbox goes live. Leave it off and the assistant simply never promises a human it can't deliver.

Honest by design. With handoff off, the assistant won't tell a user "I'll connect you to someone" — it only offers a human when you've actually staffed one. That keeps the promise real.

Next

• Governance & model policy — enable handoff and set who can answer.
• The assistant experience — where the user asks for a human.