Callab AI answers the calls the cloud left behind

Walk into the back office of a big insurer, a hospital network, or a national retailer and you will often find the same thing humming in a rack somewhere: a phone system installed years ago, wired into everything, and quietly indispensable. The software industry spent a decade telling those companies to rip it out and move to the cloud. Callab AI built its business on the opposite premise - that they never will.

Callab AI makes AI voice agents that connect directly to on-premises telephony. Instead of asking a company to migrate, it plugs into the hardware already on the wall - systems like Avaya, Cisco UCM, and Mitel - through standard SIP trunk integration. Once connected, the agents pick up inbound calls, handle the routine ones end to end, and pass the complicated ones to a human with the context attached. The company says it can automate up to 70% of inbound call volume, and that a typical deployment goes live in about a week.

The wedgeA market hiding in old hardware

The number Callab AI likes to cite is that roughly 58% of the $400 billion call center industry still runs on legacy, on-premises phone systems. For most voice AI vendors, that figure is a problem - their products assume a cloud-native stack, so more than half the market is effectively out of reach until the customer completes a migration that can take 12 to 18 months. Callab AI read the same number as an opening. The company describes its strategy as the inverse of how the rest of the market approached the problem: rather than wait for enterprises to come to the cloud, it brought the AI to them.

That framing matters because the buyer of a call center system is rarely chasing novelty. They are protecting uptime. A cloud migration is a multi-year project with real risk and a long budget cycle; an install that connects to the existing switch and works within days is something an operations lead can actually approve. Callab AI is selling into the second conversation, not the first.

There is a reason so many large organizations have not moved. On-premises phone systems are often deeply entangled with compliance rules, physical security, and internal processes built up over years. Ripping the system out means retraining staff, revalidating call flows, and accepting downtime risk during the cutover. For a bank, a hospital, or a government agency, that combination is rarely worth it for a project whose main benefit is that the phones now live somewhere else. The inertia is not laziness; it is a rational read of the trade-off. Callab AI's product design starts by accepting that read rather than arguing with it.

The productWhat the agent actually does

On a live line, a Callab AI agent handles the calls that clog a queue: routing, appointment scheduling, order status, and general customer support. It can resolve a request on its own or transfer to a human agent, carrying the conversation's context across the handoff so the caller does not have to repeat themselves. The point is not to remove the human from the loop - it is to keep the human focused on the calls that need judgment.

Language is a quieter part of the moat. Callab AI handles Arabic and its regional dialects - the messy, spoken varieties that most global voice systems render poorly - alongside English, Spanish, French, German, and Mandarin. Dialect coverage is not a checkbox; it is a way to own the calls no one else can serve well, then expand outward from there.

Dialect handling is harder than it sounds. A caller in Cairo, Casablanca, and the Gulf may all be speaking Arabic, yet the vocabulary, pace, and pronunciation shift enough to trip a system trained mostly on formal text. Getting it wrong means a caller repeats themselves, gets misrouted, and ends up more frustrated than if a human had simply answered. Callab AI treats that regional competence as a starting position rather than an afterthought, which is a natural fit given where its founders built their earlier systems. The same skill that makes the agents useful in the Middle East and North Africa is what lets the company argue it can serve a genuinely multilingual customer base elsewhere.

The foundersBuilt on infrastructure work, not a demo

Callab AI is the product of Clusterlab, a company with Tunisian roots that now operates out of San Francisco. Its two founders did not arrive at voice AI through a weekend hackathon. Chief executive Haithem Kchaou is a second-time founder who previously scaled a mobile app past 200,000 downloads and led a national AI initiative that gave more than 300,000 students free access to large language models. Chief technology officer Chehir Dhaouadi spent his career building telecom and VoIP systems across the Middle East and North Africa, and deployed local LLM instances inside a government datacenter to power that student program at national scale.

That background is the reason the company can make its central promise. Connecting AI to enterprise phone hardware is less a machine learning problem than a telephony one - SIP signaling, carrier-grade reliability, the specific quirks of a decade-old PBX. The founders had already shipped that kind of infrastructure before they pointed it at call centers. An earlier project, an AI-powered library called Elm built for Tunisia's Ministry of Higher Education, is where they say they first saw how poorly served non-Western markets were by off-the-shelf AI.

The businessWho buys, and against whom

Callab AI sells to enterprises and contact centers that run on legacy telephony - the operators most likely to have delayed a cloud move. Its reported reach spans three continents, across industries including real estate, healthcare, hospitality, retail, and debt collection, with Dunkin among the named customers. The team is small, around seven people, and the company was admitted to Y Combinator's Spring 2026 batch, reported as the first Tunisian-rooted company to make it into the accelerator.

The competitive set is crowded with cloud-native names - Sierra, Decagon, PolyAI, Retell, and others - plus the incumbent IVR menus and outsourced call center providers. Most of those competitors are fighting over the same modern, cloud-forward accounts. Callab AI's differentiation is less about a cleverer voice and more about where it can go: onto the hardware the others need the customer to abandon first. In a market this large, choosing the segment your rivals find inconvenient is a legitimate strategy.

The economics of that choice are worth spelling out. A voice agent that resolves a routine call frees a human to take the calls that actually need a person, which is where customer patience and revenue tend to sit. If a company can automate a large share of inbound volume without new hardware or a migration budget, the return does not depend on a distant payback period - it starts roughly when the install goes live. That is a different sales motion from asking a buyer to fund a platform change first and wait for value later, and it is the motion Callab AI has built around.

The readWhy the timing works

Voice AI got good enough to hold a real phone conversation only recently, and the enterprises with the most calls to automate are frequently the ones least able to change their infrastructure quickly. Callab AI sits in that seam. Its bet is that the constraint everyone else treats as a blocker - all that stubborn on-prem hardware - is actually the durable part of the market, and that being the company willing to integrate with it, in a week, in the caller's own dialect, is worth more than being one more cloud-native option.

There is risk in the position. On-prem integration is unglamorous and support-heavy, the kind of work that does not scale as cleanly as a pure API. But it is also hard to copy, which is the trade the founders appear to have made deliberately. Their entire history is infrastructure that had to work at national scale under real constraints. Pointing that at the call center market is less a pivot than a continuation.

It also fits a broader pattern in enterprise software, where the companies that win a stubborn market are often the ones willing to do the integration work everyone else avoids. Payments, data, and identity all had their version of this - a layer that spoke to old systems so the customer did not have to replace them. Callab AI is making the same argument for voice: the value is not only in the model on top but in the plumbing underneath that lets the model reach a phone line already in service. If that plumbing holds up across enough deployments, it becomes the part competitors cannot quickly reproduce.

For now, the shape of the company is clear enough: a small, engineering-led team, a specific and large market, and a wedge built out of a fact the rest of the industry would rather not deal with. Whether Callab AI becomes the default way legacy call centers adopt AI will depend on execution across a lot of messy deployments. The premise, at least, is grounded in where the phones actually are.