LOOP by mihu: The Contact Center Where AI Agents Are Colleagues

LOOP by mihu is the Mihu Contact Center: an omnichannel inbox and cloud call center where your team and your AI agents answer the same calls and the same messages, on the same screen.

What LOOP Is

Not a helpdesk with an AI agent bolted on. Not an AI agent with someone watching.

Every setup that pairs AI agents with team members breaks in the same three places. The customer explains it twice, because the bot’s transcript lives in one tool and your team’s screen in another. Your team works in three windows — the AI platform, the softphone, the CRM — and every alt-tab is a place for context to leak. And nobody can say why a conversation reached a team member, because the reason was never written down.

LOOP is built to fix that bridge. One workspace, one login, one conversation record:

1. Your team gets a full contact center: voice, WhatsApp, SMS, email, web chat, Instagram DM and Facebook Messenger in one inbox, a softphone, transfers, contacts, appointments and tasks.
2. Your AI agents — the ones you already built in Mihu — join a queue exactly like a team member does. You pick them from a list. That is the whole integration.
3. You decide the balance, queue by queue, and change it whenever you like. Outside your working hours, the AI agent takes over automatically.
4. And the work goes the other way too. A team member can hand a task to an AI agent — call this customer back, find out what they actually need, chase the document that never arrived — with a time on it if it is not for now. The agent makes the call, and what it learns is on the record before anyone opens the thread.

It is not a second product. LOOP reads and writes the same AI agents, memory, knowledge, contacts and calendar your Mihu agents already run on. There is nothing to integrate.

One Inbox, Seven Channels

Everything a floor needs, on one screen that never navigates away.

Open Inbox. Three columns: what is waiting, what you are doing, and what the AI agent knows.

1. Your queues, as rows — you only see the queues you are on. Click one and the inbox filters to it. A new hire can be live on one queue without touching the rest of the floor.
2. The conversation list — tabs for Open · Escalated · Resolved · All, and a switch from Conversations to Tasks. Every row says who is holding it, AI or team member, with a violet hairline for unread. Pin to top, mark read or unread, "unread from here".
3. The thread — voice, WhatsApp, SMS, email, web chat, Instagram and Messenger land in the same conversation. Photos, videos, voice notes, documents, locations and contact cards, both ways.

Calendar, Contacts and search open inside the same shell. Nothing reloads, so the dialer, your presence and every live subscription stay up while you look someone up. You never lose a call by opening the address book.

Anatomy of the Desk

One screen, three columns, and everything a conversation needs. This is what a team member is looking at while the call above is live.

Eight things are happening on this screen at once. Nothing on it needs a second tab.

See the desk annotated, point by point →

Three Modes, and the Switch in the Header

Not three products — three settings.

You met these on Day 2, on the queue's Automation card. Today is what they look like from your team’s chair.

Set the mode on a queue, on a channel, or on one conversation. Sales on Copilot, Support on Autopilot, VIP on Human First — one workspace, three answers. The customer never sees the switch.

The switch is in the thread header. Every change is written into the thread with a reason.

Step away and come back, and LOOP asks whether to take your conversations back from Autopilot or keep them with the AI agent. Nothing moves without you.

The Handoff, Second by Second

Four seconds, and the team member already knows everything.

This is the moment every hybrid setup gets wrong, so here is exactly what happens. A live billing call, an AI agent on the line:

What arrives with the thread is not a summary someone has to trust. It is the objects themselves:

1. The full transcript — the AI agent’s turns and your team member’s turns, in order, with who said what.
2. The contact record and everything the AI agent retrieved — the €247 charge, the account, the policy.
3. Live sentiment — a number, a label and the reason, in the customer's language.
4. The actions already taken — the AI Agent panel's Activity log is read-only, written by the platform, so no invented row can pass as an AI action.
5. The reason — written into the thread timeline, every time.

"Why did this reach a team member?" is a line in the thread, not an argument.

And it goes back: when the hard part is done, your team member hands the thread back to the AI agent for the follow-up and the reminder. The baton moves both ways.

Voice: the AI Agent on the Line With You

One live call, four states, no call transfer.

On messaging a handoff is a state change. On voice it is harder, so LOOP does it on the live call. The team member joins the same call the AI agent and the customer are on — the recording, the transcript and the conversation record never break.

You can also steer the AI agent without taking over: Say a line and the AI agent speaks it in its own voice; Transfer and the AI agent announces the transfer, then connects the line; Mute the AI agent; or End call with a goodbye message.

And you can move the call off your own screen without ending it. While it is still running, send it to another queue — billing, the workshop, whoever should have had it — or transfer the line out to another phone number entirely, a branch or a colleague’s mobile. Warm if you want to brief them first, blind if it is obvious. Either way the thread, the record and the reason go with it, so nobody restarts the conversation.

Incoming calls arrive with the brief already written: who, company, key facts, sentiment and the likely reason.

Human First on a phone number works the way a phone should: the number rings your team first. Every member of the queue rings, the first to pick up wins, and if nobody answers within the window you set (30 seconds by default) the AI agent picks up and greets. The AI agent as the fallback, not the gatekeeper.

Copilot on a live call. The AI agent hears both sides and turns the call into a script as it happens. It never speaks. What it does instead is answer on your screen while the customer is still talking. Somebody asks where their order is, and by the time they finish the sentence the order is in front of you. They ask what the refund policy actually says, and the clause is there with the article it came from. It arrives as a Say next card, grounded in your knowledge base and in whatever you have connected. The customer is talking to a person, and the person already has the answer.

This is not a voice trick. The same thing happens on every text channel, where the answer lands in the draft above the composer instead: the order status, the policy, the account detail, already written into a reply you can send, edit or throw away.

If the team member drops, the customer is not left in silence: the AI agent takes the call back after a few seconds.

Escalations You Cannot Miss

When the AI agent stops, it does not go quiet. It raises an alarm.

The hardest part of a shared desk is telling a team member, at a glance, which conversations need them. So each state looks different on purpose.

Three states, three different jobs: a red rail, an AI shimmer, an unread counter.

The conversation moves to the Escalated tab, the AI agent stops replying, and a handover strip appears in the thread with the reason. A desktop notification and a sound — an escalation is the one event we let interrupt a team member. And it reaches you even when the LOOP tab is closed: Alert me on this device is a browser push, no app to install.

Escalations are triggered by the flow rules you set on Day 2 — knowledge gap, SLA breach, negative sentiment, a customer asking for a team member, a keyword — not by the AI agent’s own judgement.

What the AI Agent Adds While a Team Member Is Typing

Only when you ask for it. In Human First, the AI agent stays silent until you press the button.

The right-hand column is what the AI agent knows: User Info, the AI Agent panel (Activity · Capabilities · Workflows), the CRM record with pipeline stages, Sessions and their labels, Tags, Notes, AI suggestions, Live Sentiment and AI Analysis.

1. Draft with Mihu AI — a button above the composer reads the conversation and writes a reply. Pick the tone — Empathetic, Professional, Friendly or the AI agent’s own suggestion — then Insert, Regenerate, Dismiss or Approve & send. It needs no configured AI agent, so a team with no AI agent gets it from day one.
2. Catch-up draft — a customer was waiting and nobody replied? The AI agent prepares the answer and asks you to approve it before Autopilot takes over.
3. AI suggestions with a confidence score — the AI agent proposes an action (book the slot, update the field, open the ticket). You Approve or Reject. Sensitive changes never happen without the tap.
4. Live Sentiment — a score, a label and the reason, updated as the conversation moves, in any language.
5. AI Analysis on every conversation — sentiment, emotion, intent, knowledge gap, whether a team member was needed, satisfaction, resolution outcome, duration and language. AI and team conversations, the same fields.
6. Ask the knowledge base — a question box over your own documents, with the source attached. Set how strict the match must be — Relaxed, Balanced or Strict — and Test this agent runs the exact retrieval it uses with a customer.
7. Live translation — the customer's language stays theirs; you read and write in yours. Keep the original visible, translate your replies, and see "Sent in Italian" on every message.

The AI agent proposes. A team member decides. Both are logged.

Try this: put one queue on Copilot for a day. Count how many drafts your team sent unchanged. That number is the work you were paying your team to type.

Contacts, Appointments and Tasks

One record, one calendar, three authors.

This is the test of whether AI and your team really share a desk: can both write to the same record and the same calendar? Yes — with a marker saying which one did.

One contact record. In Autopilot the AI agent writes it. In Copilot it proposes and your team member confirms. In Human First your team member owns it outright. Custom fields, tags and pipeline stages are yours, so a conversation can move a deal. Sensitive field changes queue for approval. Every past conversation across every channel sits on the contact, with its analysis. An edit lands the moment you make it: the colleague who picks the thread up next sees it, your reporting counts it, and the AI agent answering the following message already knows.

The AI agent fills it from the conversation. Nobody types up a record after a call. The AI agent reads the thread as it happens and pulls out the fields it can see — the company, the vehicle, the order number, the reason they got in touch — and puts each one where it belongs. What happens next is yours to set. The fields can queue as a suggestion with a confidence score, so your team member glances at them and approves with one tap, or they can go straight through to your CRM with nobody touching them at all.

You set that per field, which is the part that matters. A corrected address can write itself; a payment term, a discount or a cancellation waits for a person. Either way the field is marked with where it came from, so a record nobody typed is still a record you can audit.

One calendar. Either side can book, move or cancel, and a team member can override any of it. You decide how much authority the AI agent has on the schedule — always ask for approval, auto-approve minor changes, or full autonomy — and whether a team member confirms its bookings.

Tasks — the outbound queue. Most of a contact center's value is outbound, and most outbound dies in a spreadsheet. Flip the inbox to Tasks and you have the calls and messages that should happen today: new leads, callbacks, reminders, win-backs, complaint follow-ups. Each opens with the reason it exists, the contact record and a suggested opening line. Create one for me, or for an AI agent — pick the AI agent, the contact, the channel and the due time, and the AI agent makes the call.

A task does not have to be for a person, and it does not have to be for now. Hand one to an AI agent with a time on it — call this customer in three days, send that WhatsApp tomorrow morning, email the quote on Monday — and the agent does it when the clock says so, in your customer’s language, with the whole history already in front of it. It opens as an ordinary conversation on the desk, so anyone on the queue can step in mid-call.

And it can finish the job on its own. If the call ends with a date agreed, the AI agent books it straight into the calendar, sends the confirmation, and sets the reminder that chases the no-show — no ticket, no handover, no person retyping a slot that was already spoken out loud.

Languages, Presence and the Things That Ring

The parts nobody demos and everybody needs by week two.

1. 16 interface languages — Türkçe, English, Italiano, Español, Nederlands, Français, Deutsch, Português, Українська, Русский, Polski, Română, Български, Čeština, العربية, עברית — with full right-to-left layouts for Arabic and Hebrew. Live translation across 40+.
2. Presence that means something — Available, Working, On a call, Lunch, Be right back, Out of office. Go idle and LOOP notices; set Lunch and it never overrides you. When you are not available, the composer locks: replies are paused until you are back.
3. A visible activity log on your own account — every request made with it, newest first. Team management, roles and access logs stay where you set them on Day 2.
4. Your own domain and logo on the desk. The same workspace runs under a partner's brand, out of the box.

What LOOP Costs

A monthly plan, free seats inside it, and usage you can read off a table.

Contact center software is normally sold by the chair. You count your team members, multiply by a monthly licence, and every new hire is a line item before they have answered anything. LOOP does not put a price on a person. You pick a plan, and everyone you put on the desk is free inside it.

The plan is what sets the ceilings: how many people you can seat, how many AI agents you can run, how many channels one of them may work, how many inbound and outbound email addresses you get, and what you are allowed to train an agent on. Seats stay free at every level, so growing a team is never the thing that moves your bill.

What Your Team Costs, and What the AI Costs

Inside the plan, only two things are metered: the minutes your team spends on the phone, and the work you hand to the AI agent.

Neither of those two moves with your plan. Freemium pays what Enterprise pays.

Everything the AI agent does is metered instead, per message or per minute, and what it costs is based on your plan. The bigger the plan, the less each AI message and each AI minute costs, so the busier you get the cheaper the work the AI agent is doing for you. WhatsApp and the other Meta channels, SMS and calls also carry their own carrier rates, and calls depend on where you are dialling. Every rate, plan by plan, is on the plans page.

That distinction is worth a second read, because it changes what you can do with the product. A team that has switched no AI agent on pays its plan and its phone minutes, and nothing else. Seven channels, the softphone, transfers, contacts, appointments, tasks and the outbound queue, with every message free.

So the AI agent is never the price of admission. You can put your team on the desk first, work it for a month, and then turn an AI agent on for the one queue where the repetition is worst. That queue starts metering. The others carry on as they were.

A worked example. Seven people on Starter, ten thousand minutes on the phone last month, every message answered by hand.

Seven people, ten thousand minutes, seven channels. A hundred and ten euro.

Nothing in that bill was counted per person, so the seventh hire cost exactly what the first one did. Two things move it: outgrowing your plan, and deciding the AI agent should take some of the work.

1. Seats are free inside your plan. How many you get is set by the plan you are on, and the larger plans stop counting altogether.
2. Numbers you buy in-app, live in minutes. Connect WhatsApp, Instagram, Messenger, email, SMS and the web chat widget the same way.
3. Two-factor authentication, GDPR and HIPAA flags per AI agent, recordings you control.

Try this: register, buy a number, open the Day 2 queue, and put one AI agent in it beside one team member. Take a call. Press Take over. Press Hand back to AI agent. That is the whole product, and it takes an afternoon.

Frequently Asked Questions