Human Handoff

Run a live desk. Step in when AI needs backup.

Airgentic was built as an AI-first service — not a contact centre with a chatbot bolted on. Human Handoff is the live desk around those conversations: My chats, Queued, and Monitoring, hop without notifying the visitor, and take over when judgment matters. AI handles the routine. Humans stay in control.

Contact us for a live demo
Human Handoff

The problem:Why this matters

Even strong AI needs an escalation path. If AI is customer-facing, handoff cannot be an afterthought.

  • Some customers want reassurance from a real person
  • Some questions need human judgment, not just retrieval
  • Some conversations show frustration before they become complaints
  • Without live visibility, teams only discover failures after the moment has passed
  • A contact-centre product with a chatbot bolted on still treats AI as an afterthought

The solution

Airgentic Human Handoff keeps humans in the loop without slowing down the AI.

Because the product started as conversational AI, the desk is the human path around those chats — not a separate contact-centre stack. Three columns: inbox, transcript, and visitor context. Live chats sit in My chats, Queued, or Monitoring. Search, filter, and sort the list; queued chats turn amber then red as someone waits. Switching rows only changes what you see. Agents take over to pause the AI and reply — from an AI draft, a saved reply, or their own words — then hand back, transfer, or return to the queue. Internal notes stay with the team. Staffed hours, auto-assign up to ten, and an Agent role complete the desk. Three modes: AI Only, Human-Supported AI, and Human Only.

Why it works

Live inbox

My chats, Queued, and Monitoring update automatically — no refresh

Hop without notifying

switching rows only changes the screen; the visitor is not told you looked

Search, filter, and sort

find a chat by question, visitor, service, or tag; sort by longest waiting or newest

Wait-time colours

queued chats turn amber, then red, so long waits are obvious

Recent 48 hours

closed conversations stay findable, including threads a visitor reset

Instant take-over

pause the AI and reply as a human in the same thread

Staffed hours

Human-Supported during the day, AI Only after hours and on closed dates

Auto-assign

waiting enquiries fill My chats up to ten; a toast and unread mark, never a stolen screen

Assistive answers

generate an AI draft, revise it, then send it as a human reply

Saved replies

insert a prepared answer from Replies or by typing /

Agent role

let support staff work the inbox without changing service settings

Visitor context

captured attributes and a conversation summary sit in the right column so the rep does not start from scratch

How it works

Work the inbox

See every live chat, find the one you need, and hop without the visitor knowing.

Know what needs you

Attention flags, wait-time colours, and alerts — so you don’t have to stare at the screen.

  • Spot conversations that need attentionThumbs down, a non-answer, negative sentiment, or an explicit ask for a human — highlighted on the card, counted on a banner and the browser tab.
  • See how long they have been waitingQueued chats turn amber, then red. Admins set the times per service. A triage signal, not an automatic assignment.
  • Auto-assign waiting enquiriesAvailable agents receive the oldest waiting chat into My chats, up to ten. A toast and unread mark — never a stolen screen.
  • Stay alerted without babysitting the screenTab count, toasts, audio, and optional desktop notifications — even if you are elsewhere in the admin console.

Take over and reply

Pause the AI, answer as a human, then hand back, transfer, or return to the queue.

  • Take over when you need to replyClaim from Queued or Monitoring. The AI pauses, and in Human-Supported the visitor is told a human has joined.
  • Generate, revise, and sendGenerate an AI answer, insert a saved reply, or write from scratch. Citations stay in the admin pane.
  • Leave an internal noteFor colleagues on the thread. The customer, the AI, and Insights never see it.
  • Hand back, transfer, or return to queueHand back resumes the bot (they are told). Transfer names a colleague. Return to queue keeps the AI paused, with no visitor message.

Run the desk

Hours, overnight coverage, and recently closed threads.

  • Follow staffed hoursWeekday hours and closed dates, with timed overrides when the desk needs to stay open.
  • Review Recent for 48 hoursClosed conversations — including threads a visitor reset — stay available as read-only for two days.

Service modes for every operational model

Human Handoff adapts to the way your team wants to work.

This lets you run different services with different levels of autonomy, from full automation to full manual control.

AI Only

the AI handles all conversations without intervention

Human-Supported AI

the AI leads, while agents can monitor, step in, and handle escalations

Human Only

automated responses are disabled until a human replies

Built for a staffed desk

Takeover is the start. The desk covers an inbox you can hop through, hours, assignment, drafting, saved replies, wait-time colours, and who is allowed to change settings.

Staffed hours

Choose weekday hours and a timezone. Typical setup is Human-Supported AI while the desk is open, and AI Only overnight, on weekends, and on closed dates. Admins can override the live mode until the next scheduled change without rewriting the hours.

Enquiry auto-assign

Optional per service. Human Only queues every new conversation; Human-Supported queues only chats that need a human. Agents join the pool when Human Handoff is open and Available is on. Auto-assign fills My chats up to ten concurrent claims and never steals the screen you are working.

Assistive AI answers

Generate the same answer the customer would have seen, edit it, and send it as a human-agent reply. Clear and write your own if you prefer. Generated drafts count toward usage; typed replies do not.

Saved replies

Admins write ready-made answers with placeholders for visitor name, agent name, and service. Agents insert them from Replies or by typing /.

Wait-time colours

Queued chats turn amber, then red, as the customer waits. Set the times per service so the queue shows what needs you first.

Search, filter, and sort

Find a chat by question, visitor, service, or tag. Filter to needs-attention only. Sort by longest waiting or newest.

Agent role

Give live-support staff an Agent login. They can work the inbox and Insights — they cannot change modes, hours, auto-assign, or other configuration.

Where Human Handoff shines

Customer service teams

intervene on frustrated or unresolved conversations before they turn into complaints

Local government

give residents a human path for sensitive or complex service enquiries

Technical support

step in when troubleshooting requires interpretation or reassurance

Sales conversations

rescue high-intent prospects when the AI reaches the edge of its scope

High-risk communications

keep humans close to conversations involving policy, exceptions, or unusual cases

What human agents can see

Each live conversation card gives enough context to act quickly:

  • Whether the AI is active or paused
  • Which specialist agent is handling the chat
  • Visitor name and city
  • A visitor label when Visitor Profiles has captured a name or other identifying fields
  • The opening question
  • Start time and message count
  • How long they have been waiting — on queued chats, and on your chats with an unread customer message
  • Conversation tags the claimed agent has applied

The workspace is three columns: inbox, transcript, and visitor context. Selecting a row does not notify the visitor. Inside the transcript, agents see clear separation between user messages, AI responses, human messages, system notifications, and internal notes. Each message shows a time when it is known, with day dividers when a conversation spans more than one day. When Visitor Profiles is enabled, the right column shows structured attributes (not a paragraph to interpret) and a short AI summary of why they came and what is still open.

After taking over they can generate an AI draft, insert a saved reply, leave an internal note, or write from scratch.

Governance & control

Human Handoff is a staffed desk with controlled intervention — not an unsupervised chat dump.

Claiming controls

if another agent has already claimed a conversation, that is clearly shown

Explicit AI pause

the AI stays paused until you Hand back to AI; hopping and returning to queue do not unpause it

Hand back, transfer, return

three distinct actions, with a visitor message only when you intend one

Mode-based control

set services to AI-led, human-supported, or human-only

Staffed hours

weekly hours and closed dates, with timed overrides when you need them

Auto-assign

admins turn the pull-queue on or off per service; claims fill My chats up to ten

Wait-time colours

amber then red as queued chats wait; a triage signal, not an automatic assignment

Agent role

live-support staff can work the desk without changing configuration

Internal notes

team-only notes on the thread; the visitor and the AI never see them

Operational visibility

a need-attention banner and a count on the browser tab; My chats, Queued, and Monitoring show their own counts

Example use cases

  • Unsatisfactory AI answer — A customer asks for help, receives an unsatisfactory answer, and gives a thumbs down. The conversation is flagged instantly, lands in Queued, and a support agent steps in — optionally after generating an AI draft they can revise before sending.
  • Sensitive resident enquiry — A resident asks a sensitive question about a council matter. The service runs in Human-Supported AI mode so staff can intervene whenever needed.
  • Technical edge case — A troubleshooting conversation reaches an edge case. An agent takes over, resolves the issue, then hands the conversation back to the AI for follow-up.
  • Staffed campus hours — A university runs Human-Supported 09:00–17:00 and AI Only overnight and on closed dates. A timed override covers an evening open day without rewriting the weekly hours.
  • Busy Human Only queue — Auto-assign fills My chats for the next Available agent (up to ten), so nobody has to pick from a pile of cards. The screen they are already working is not stolen.
  • Long wait in the queue — Queued chats turn amber then red. An agent filters to needs-attention, takes the oldest red chat, inserts a saved reply, and leaves an internal note for the next shift.
  • Hop between several chats — An advisor holds a few My chats at once, switches rows to check a second visitor, then returns — neither person is notified until the advisor takes over or sends.
  • Service disruption — During an incident, a team sets a service to Human Only so all responses are handled manually until things stabilise.
  • Recruiter pickup — An advisor takes over a future-student chat. The right-column visitor profile already shows campus, programme interest, and email, so they do not ask for details the person already gave the AI.
  • Recent follow-up — A visitor closed the widget. The thread stays in Recent for 48 hours, so the advisor can still see what was said.

The result

Human Handoff gives you the speed benefits of AI without giving up accountability, empathy, or operational control.

Customers get fast answers when AI can help, and real people when it matters. Your team gets a live inbox designed around those conversations — hop, search, take over, transfer, and hand back — because Airgentic was built as an AI-first service, not a contact centre with a chatbot bolted on.

FAQs

Does Human Handoff require agents to refresh the page?

No. The Human Handoff screen maintains a live connection to the conversation server and updates automatically.

What do the connection status indicators mean?

A green dot means you’re connected and receiving live updates. Yellow means the page is connecting. Red means the connection was lost—the page attempts to reconnect automatically.

How do we see that a conversation needs attention?

A red banner above the inbox shows how many chats need a person, and the same count appears on the browser tab (for example “(3) Human Handoff”). Attention flags also highlight those cards. My chats, Queued, and Monitoring each show their own counts.

What are My chats, Queued, and Monitoring?

My chats are conversations assigned to you — you can hold several at once. Queued is unclaimed chats that need a human now (an escalation, Human Only, or a paused chat returned to the queue). Monitoring is live chats the AI is still answering; a new Human-Supported conversation starts there. Recent is a separate tab for closed conversations from the last 48 hours.

Does looking at a chat notify the visitor?

No. Selecting a row only changes what you see on screen. It does not claim or release the conversation, and the visitor is not told you looked. They are notified when you Take Over (in Human-Supported), Transfer to a named colleague, or Hand back to AI.

Can we search or filter the inbox?

Yes. Search finds chats by question, visitor name, location, service, or tag. Filter by service, tag, or needs-attention only, and sort by longest waiting or newest. Hidden chats still count in the need-attention banner and tab badge, so a filtered-out escalation is never silently lost.

How do we see chats that have been waiting too long?

Queued chats turn amber, then red, as the customer waits. Admins set those times per service (defaults are two minutes and five). The colours are a triage signal — they do not auto-assign or page anyone.

Can agents use saved replies?

Yes. After taking over, insert a saved reply from Replies or by typing “/”. Placeholders such as visitor name, agent name, and service name are filled in on insert. Admins write and edit the list in Desk settings.

Can agents leave a note the customer cannot see?

Yes. After taking over, Note switches the composer to an internal note. The customer, the AI, and Insights never see it — it is for the next agent on that thread.

What triggers a handoff or attention flag?

A conversation can be flagged when the customer gives a thumbs down, the AI cannot answer (non-answer), negative sentiment is detected, or the customer explicitly requests a human.

Can multiple agents try to take over the same conversation?

Yes, but once a conversation is claimed, other agents are shown who owns it and the Take Over button is disabled until it’s transferred, returned to the queue, or handed back to the AI.

What’s the difference between Hand back to AI, Transfer, and Return to queue?

Hand back to AI resumes the bot from the next visitor message and tells them. Transfer assigns a named colleague and tells the visitor once (for example “Archie is now helping you.”). Return to queue drops your claim, keeps the AI paused, and does not notify the visitor — the chat waits in Queued for the next agent.

Can the AI resume after a human steps in?

Yes. An agent can Hand back to AI at any time, and the AI will continue from the next customer message. A system message is shown to the customer. Hopping between chats, auto-assign, and returning a chat to the queue do not unpause the AI.

Can we disable AI completely for a service?

Yes. Services can be set to Human Only, which disables automated responses until a human replies. You can also use AI Only (no handoff) or Human-Supported AI (standard handoff mode).

Will I get notified if I’m on another admin page?

Yes. Toast notifications and a sound alert appear even if you’re elsewhere in the admin console. The browser tab shows a count whenever chats need attention, and you can turn on desktop notifications (the bell) so the OS alerts you when the tab is in the background. Auto-assign still requires the Human Handoff page to be open with Available on.

What information do agents see for each conversation?

Each card shows status (AI Active or Paused), the specialist agent type, visitor name and city, opening question preview, start time, message count, and wait time where it applies. Selecting a row opens the full colour-coded transcript in the centre column, with visitor context on the right. Messages show a time when it is known.

Where do closed conversations go?

The Recent tab keeps closed conversations for 48 hours, including threads a visitor reset. They are read-only — no composer, Take Over, or transfer.

Can we staff Human Handoff only during office hours?

Yes. Admins can set weekly hours and closed dates per service. A typical setup is Human-Supported AI while the desk is open, and AI Only overnight, on weekends, and on holidays. You can still override the live mode until the next scheduled change, until a chosen time, or until you turn the schedule off.

How does auto-assign decide who gets a conversation?

It is optional and off by default. Agents join the pool when they have Human Handoff open and Available is on. In Human Only every new chat is queued; in Human-Supported only conversations that need a human are queued. The oldest waiting conversation fills the next idle agent’s My chats, up to ten concurrent claims. Assignment never opens the transcript — you get a toast and an unread mark. If nobody is free, it stays in Queued until someone has capacity. Manual Take Over from Queued can exceed the cap.

Can agents generate an AI answer without sending it to the customer?

Yes. After taking over, Generate AI Answer runs the same agent pipeline the customer would have seen and places the markdown in the New Message box, with source citations underneath for reference. Edit it, clear it, or write your own, then Send. The customer sees a human-agent reply. Generated drafts count toward usage; replies you type yourself do not.

Can we give someone the desk without giving them admin access?

Yes. The Agent role can work Human Handoff and Insights — including saved replies and internal notes. Agents cannot change service modes, staffed hours, auto-assign, wait-time colours, saved replies, or other configuration.

Do agents see more than the chat transcript?

Yes. When Visitor Profiles is enabled, the right column shows the captured attributes and a short conversation summary. Conversation cards can also show a visitor label (for example a name and programme). Without Visitor Profiles, agents still see the live transcript, location, and tags.

Get started

Ready to see Airgentic on your site?

Still have questions?

Ask our AI — get cited answers about the live inbox, wait-time colours, saved replies, and handing a chat to an agent.