> ## Documentation Index > Fetch the complete documentation index at: https://docs.replybubble.com/llms.txt > Use this file to discover all available pages before exploring further. # Human Handoff > Let visitors escalate from your AI chatbot to a live team member Human Handoff lets a visitor move from your AI chatbot to a real team member without leaving the chat widget. The AI can also trigger a handoff on its own when it can't answer a question or the situation calls for human judgment. *** ## Enabling Human Handoff Go to **Dashboard > Settings** and toggle **Human handoff** ("Allow visitors to request a human agent from the widget"). It's off by default. When enabled, two things change: * The widget shows a **Talk to a person** button next to the message composer. * The AI is given a `transfer_to_human` tool it can call on its own -- the AI is instructed to send the visitor a brief message first, then hand off, whenever the visitor explicitly asks for a human, the AI can't answer the question, or the situation needs human judgment (complaints, billing disputes, etc.). When Human handoff is off, the AI instead gets a `collect_contact_info` tool: it triggers your contact capture form so the visitor can leave their details for a follow-up, since there's no live agent to transfer to. *** ## How Visitors Request a Human There are two paths into a handoff: 1. **Manual** -- the visitor clicks **Talk to a person** in the widget. 2. **AI-initiated** -- the AI calls `transfer_to_human` with a reason (`user_requested`, `cannot_answer`, or `complex_issue`) and a short summary for the agent. Either way, the conversation is escalated and the widget switches into a waiting state while it looks for an available agent. *** ## What Visitors See The widget shows a status banner that changes as the handoff progresses: | Time waiting | What appears | | ------------ | ------------------------------------------------------------------------------------------------------- | | 0-30s | "Connecting you with a team member..." with a **Stay with AI** button to cancel | | 30-60s | "Still looking for an available agent..." with **Leave a message instead** and **Stay with AI** options | | 60s+ | "No agents available right now" and a leave-a-message form (email + optional message) | The visitor can keep typing while waiting -- a note reminds them "Your messages will be seen by the next available agent." If they cancel, the AI resumes immediately. If no agent claims the conversation within a minute, the widget offers a leave-a-message form. Submitting it sends the visitor's email and message and confirms with "Message sent. We'll reply by email." This is separate from [Offline Mode](/customization#offline-mode), which only applies when your AI provider itself is unavailable. Once an agent claims the conversation, the banner switches to "**\[agent name]** is here to help." When the agent releases the conversation, the widget shows "You're now chatting with AI again" and the banner clears itself a few seconds later. *** ## Business Hours Interaction If you've configured a business hours schedule, the widget only offers Human handoff during those hours -- the `humanHandoffEnabled` flag sent to the widget is `true` only when the workspace setting is on **and** the current time falls within your configured schedule. Outside business hours, the widget falls back to the AI-only experience (or your configured offline behavior) even if Human handoff is enabled in settings. *** ## How Agents Take Over (Inbox) Escalated conversations appear in the **Inbox** with an **Escalation Context** card at the top of the thread showing: * The visitor's name and email, if known * How long they've been waiting * Why the conversation was escalated ("Visitor requested human agent" or "AI detected need for human help") * An AI-generated summary of the conversation for context, when one is available An unclaimed escalation shows a **Claim & Reply** button. Any conversation can also be claimed from the chat panel header: * **Claim** pauses AI responses and assigns the conversation to you. * **Release** hands it back -- AI responses resume. You can also assign a conversation to a specific teammate, or use auto-assign, from the agent picker in the chat panel header. *** ## AI-Resume Behavior While an agent is assigned, the AI does not respond -- only the human agent's replies reach the visitor. When the agent releases the conversation (or unassigns), the widget shows a system message that the agent has left, briefly displays "You're now chatting with AI again," and the AI resumes handling the conversation. *** ## Escalation Effects on the Conversation Escalating a conversation: * Sets its status to reflect it needs attention and marks the resolution as escalated to a human * Records an `escalatedAt` timestamp and, when available, the escalation reason * Generates a short AI summary of the recent transcript in the background (2-3 sentences covering what the visitor wants, what the AI already tried, and what they likely need next) if the AI didn't already provide one when it called `transfer_to_human` A conversation can only be escalated once -- if it's already been escalated, a repeat request is a no-op. The escalation timestamp is preserved even after the conversation is later closed, so escalation counts stay accurate in [Conversations](/conversations) and your analytics reporting. *** ## Tips * **Review the escalation reason before replying.** "AI detected need for human help" usually means the AI hit a real knowledge gap -- worth checking whether your knowledge base needs an update. * **Claim quickly.** The widget only waits 60 seconds before offering the visitor a leave-a-message fallback, so an idle Inbox means missed live conversations. * **Release when you're done**, don't just navigate away -- releasing is what tells the AI to resume responding to the visitor.