Human only
You, unaided
Cloud call center
Numbers, queues, routing, transfers and quality scoring — the whole operation, running in a browser tab. No hardware to rack, no phone system to migrate, no per-seat licence.
Everything included
This replaces the stack, it doesn't sit on top of it. Get a number, or keep the phone system you already own and add Mihu as an extension. Connect the channels, set the routing, run inbound and outbound, keep the records, score the quality. One platform, one bill, one login for your whole team.
Coverage
Calls, WhatsApp and email land in one queue during office hours, and your people take them first. The moment nobody can pick up — after six, on a Sunday, or when the line's already full — Mihu takes over instead of the customer hearing a ring-out.
AI agents
Everything above works with a team of people and nothing else, and many customers run it exactly that way. You can also add AI agents to the same queue, handling the same channels, visible in the same list. What never changes is who decides: the AI takes volume, you keep control of anything that matters.
Mihu can answer a conversation on its own. What keeps it safe is that you decide in advance what it's allowed to do, you can watch it happening in the same queue your team already works in, and you can take any conversation off it in one click. Nothing it does is hidden, and nothing sensitive happens without a person saying yes.
It doesn't mean handing customers to a black box and hoping. Every conversation Mihu handles sits in the same list as the ones your team is handling, marked so you can see at a glance who has it. When Mihu hands one back, it says so in the thread — and the person picking it up has already read everything.
What stops most teams putting AI in front of customers is not capability. It is control.
An AI agent that is right 95% of the time and unpredictable the rest is not safe to deploy, because the remaining 5% is where the damage happens. The question that decides whether it ships is not how capable it is. It is how much of it you still control.
Why teams switch
Until recently you had two options: an automated system customers try to get past, or enough staff to answer every call yourself. One damages the experience, the other costs more every time volume grows. Running both in one queue removes the trade-off.
Human · Co-pilot · Autopilot
Every conversation your team takes — a call, a WhatsApp thread, an email, a DM — sits in one shared queue with a control at the top. That control has three positions, it belongs to the conversation rather than the company, and by default it starts wherever you put it. Plenty of teams leave it on the first one.
You answer. Mihu stays silent — no drafts, no suggestions, no interruptions — but the whole thread is already in front of you, from every channel the caller has ever used.
Mihu looks up the order, checks the policy and writes the reply. You read it, change a word if you want, and press send. Nothing reaches the customer without a person choosing to send it.
Mihu handles it end to end — answers, books, refunds, updates the record — inside the rules you set. And the second something falls outside them, it stops and hands the conversation to a person.
You, unaided
Mihu drafts · you send
Mihu runs it · hands back to you
Guardrails and permissions
Control comes from three settings that work together: the boundaries an AI agent cannot cross, the conditions that force a handover to a person, and the approvals you require before anything sensitive happens. You configure all three by writing them in plain language — no scripting.
The non-negotiables, applied to every conversation no matter how the customer phrases it or how hard they push. Always verify identity before discussing an account. Never quote a final discount. Never give medical or legal advice. The agent is free to be genuinely helpful inside a space you drew — and simply cannot step outside it.
Some conversations should go to a person, and the skill is spotting them early. Frustration. A complaint. Anyone who asks for a human. A high-value customer, a sensitive subject, anything off-script. When a trigger fires, the thread moves to the right person with the whole history attached.
Require a person to approve before anything sensitive happens — a credit, a high-value booking, a particular message. Watch conversations live and step in instantly. Let the agent pause and ask a human for a decision rather than guess. Then review how it went and tighten the rules.
Escalation
When an escalation trigger fires — frustration, a repeat complaint, a request for a person — the conversation moves to the right agent with the full transcript attached. The customer does not repeat anything, and there is no hold music in between.
How we compare
The common setup is a contact center built for people plus a separate AI layer, sold together and connected by an integration. It works until the join fails — the call that never reached the record, the escalation that arrived without history, the bot that did not know a person had already promised something.
Who uses it
The same platform covers both, because the amount of work you hand to AI is a setting rather than a plan you commit to.
Take on twice the volume without twice the headcount. Put the routine on Autopilot, keep your experienced people for the conversations that need them, and score every interaction instead of a sample. One system across every channel, language and region — and a dial you can turn back the moment quality slips.
Never leave a call or a message unanswered, including at 9pm on a Sunday. Sound like a full team when it's three of you. Capture every lead and follow it up automatically — and stay personally in the loop on anything that matters, because you're the one who set the rules.
FAQ
No. A chatbot answers questions from a script. Mihu takes actions in your real systems — it looks up the order, books the appointment, issues the replacement, updates the record — and it does that across phone calls as well as messages. The part that makes it different from a bot is that a person can stop it, correct it, or take the conversation over at any point.
The opposite. The point of keeping a human in the loop is to spend your time well, not to have less of it. The agent absorbs the hundreds of routine conversations so you are free for the handful that genuinely need judgment, empathy or authority — which is also the work people would rather be doing.
Guard rules. You write down what the agent must always do and must never do — never quote a final discount, never give medical advice, always verify identity first — and those apply to every conversation regardless of how the customer phrases the question. If someone pushes at the edge of a rule, the conversation goes to a person instead.
No. You describe what you want in ordinary sentences — "if a customer sounds frustrated or asks for a person, hand off to the duty manager with the full history" — and Mihu applies it. Same for building an agent, changing a rule or pulling a report. If you can explain your policy to a new starter, you can configure it.
By watching. Every conversation is scored for sentiment, intent and whether it actually resolved, so you can see how the agent handles a given type of request before you let it handle that type alone. Most teams start on Co-pilot, watch for a few weeks, and move categories to Autopilot one at a time as the evidence comes in.
Nothing they'd notice, except the answer getting better. It's the same thread on the same screen — no transfer, no new reference number, no repeating the story. If a person takes over from the AI, they've already read everything that was said before they arrived.
No. You can get a number inside Mihu and be taking calls in minutes, or keep the phone system you've already invested in and add Mihu to it as an extension over SIP. Both are supported, and you don't have to decide on day one.
You look at the queue. Every conversation is labelled with who has it — your own name on the ones you're handling, "Mihu AI handling" on the ones it's working, "Mihu AI replying" while it types, and "Escalated" on anything it has pushed to a person. It's the same list your team already works from, so there's no separate AI dashboard to go and check.
A number, a queue and a login is all it takes to start. Add channels, campaigns, reporting and as much AI capacity as you want, whenever you want — without changing platforms.
Numbers · Queues · Routing · Channels · Campaigns · CRM · QA · €30 free credit