Human handoff

When AI reaches its limit, a human steps in

Some conversations need a person. Your agent recognizes the moment, escalates to a real inbox, and a teammate picks up the thread with full context — then resolves it back to the AI when they're done. The user never has to repeat themselves.

What the handoff flow gives you

  • Escalate on genuine intent

    When a user asks for a human, the agent triggers a handoff instead of guessing an answer. A cheap classification guard double-checks the request first, so a stray phrase never drags a conversation out of automation by mistake.

  • A real-time queue

    Every waiting conversation lands in a shared inbox with status, channel, and an AI-written summary of what's happened so far. Your team sees who's waiting and for how long — and presence pings show which teammates are online to take them.

  • Live chat with full context

    When a teammate accepts, they pick up the whole conversation — not a cold transfer. Messages flow in real time both ways, so the reply reaches the user on whatever channel they started on, and the user never repeats themselves.

  • Every channel, one inbox

    Telegram, Discord, Slack, Meta, WhatsApp, Google Chat, and the web widget all route into the same place. Your team answers from one screen; each reply is delivered back on the user's own channel automatically.

  • Nobody left waiting

    If no one picks up within a few minutes, the team gets an email reminder; if a session waits too long, it auto-expires gracefully and returns the user to the AI with a clear message instead of leaving them hanging.

  • Scoped, controlled, and tier-aware

    Every queue and message is scoped to your organization, with role-based access for who can view versus accept and reply. Handoff can be turned on per agent, and availability follows your plan tier.

From AI to human and back

A handoff is a round trip, not a dead end. The agent escalates, a teammate takes over, and resolving returns the conversation to automation — with the user carried along on their own channel the whole way.

  1. User asks for a human

    Mid-conversation, the user requests a person. The agent recognizes the intent rather than forcing an answer it isn't confident about.

  2. Guard & escalate

    A lightweight classification check confirms the request is genuine, then the conversation is escalated and a queued session is created with an AI summary attached.

  3. Queue & notify

    The session appears in the team inbox and an in-app notification fires instantly; if no one accepts within a few minutes, an email reminder follows.

  4. Accept & chat live

    A teammate accepts and chats with the user in real time. Replies are delivered back on the user's original channel — Telegram, WhatsApp, the widget, wherever they started.

  5. Resolve back to AI

    When the issue is handled, the teammate resolves the session and the conversation returns to the AI agent — or auto-expires gracefully if it was never picked up.

Explore other features