In 2017, Stakh Vozniak sat in a basement in Iowa and called truckers. Sixty calls a day. Most went nowhere. Some ended with laughter; others with a click. He had arrived from Ukraine for a startup accelerator with six months of runway, weekly English grammar lessons and a product no one was buying. It is a peculiar origin for a company that now sells tireless, multilingual digital workers. It is also the only origin that makes the product make sense.
The freight in five
- Cargofy automates freight work for carriers, shippers, brokers and 3PLs.
- Its agents search, call, bid, book, track and process documents inside tools customers already use.
- Small operators can pay per booking; larger customers pay subscriptions, usage fees or custom contracts.
- The original owner-operator app was useful, but 2022 forced the company toward a broader AI product.
- The model works best when a customer can define a painful workflow and stays involved in deployment.
Those early calls were not a heroic montage. They were field research with bad acoustics. Vozniak learned what a carrier wanted to hear in the first eight seconds, which objections were real and which were polite exits. Years later, Cargofy would encode those lessons into agents that call thousands of carriers, in multiple languages, without losing patience or needing lunch.
“You can't outsource the first 60 calls a day.”Stakh Vozniak, co-founder and CEO
The product that worked - and still had to change
The first Cargofy was a mobile assistant for owner-operators: find a load, book it, coordinate the trip, track the delivery and manage the money around it. That product still exists. The Android app advertises instant booking, an active-trip assistant called KAI, truck-specific locations, fuel discounts and financial tools. Google Play shows more than 10,000 downloads.
It solved an obvious problem. Independent truckers often stitch together work from multiple load boards while trying to avoid empty miles. Cargofy aggregated the hunt and made the phone feel more like a dispatch desk. The awkward part, Vozniak later admitted, was that it worked without being especially original. The team had wanted to build with AI almost from day one, but the language models available in 2018 were too crude for real freight conversations.
Then 2022 rearranged both the market and the technology. Russia's full-scale invasion of Ukraine hit a customer model that leaned heavily on recent immigrants to the United States, including many Russians. At roughly the same time, modern language models made the old ambition plausible. Cargofy returned to the idea of a worker that could call, collect information and take action. The team built much of the voice infrastructure itself.
Software that clicks the buttons
Most freight software is a very organized spectator. It shows a dispatcher a dashboard, places a rate beside a lane and waits for a human to do something. Cargofy wants the software to do the work: source carriers, request quotes, negotiate rates, book loads, coordinate drivers, watch tracking feeds and handle paperwork. It reaches people through the channels freight already uses - phone, email and WhatsApp - and sits on top of existing TMS platforms and load boards.
One job, five movements
That difference sounds small until volume enters the room. A tender manager may email hundreds of approved carriers and receive only a handful of bids. An agent can follow up with every carrier before the tender closes and repeat the call when the leading rate changes. A dispatcher can monitor a few load boards for a few trucks. An agent can monitor many feeds continuously and make first contact while the load is still available.
Cargofy's cleanest trust-building example is almost comically specific. A fleet with 800 trucks wanted a call placed whenever a driver stopped at a truck stop and left the engine running for five minutes. The agent watched the GPS feed and called with the company policy. Cargofy says the fleet saved about $80,000 a month in fuel. No grand reinvention was required. The agent simply noticed every instance, including the one at 2 a.m.
Eight price tags, because freight has eight wallets
Cargofy does not really have one business model. It has three customer families - shippers, carriers and forwarders - and, according to Vozniak, eight commercial combinations among them. For a one-truck owner-operator, the subscription can be zero until a load is booked; Cargofy then takes a percentage. A shipper without a carrier network can compare prices for free and pay on a booking. Mid-market buyers pay a subscription with usage on top. Large enterprises negotiate annual contracts.
The enterprise formula is revealing. The company estimates the hours and money it can save, then aims to charge roughly a tenth of that value. An investor who taught pricing at Stanford nudged the team away from calculating a tidy software margin and toward measuring business impact. If an agent saves a client $1 million a month, the relevant question is not whether the server bill supports a $5,000 subscription.
What failed first was the introduction
The early digital workers had abstract names such as Jane and Alice. Customers complained when “the AI” did something wrong, even when Cargofy's review suggested a human had made the error. The product was technically capable and socially irritating. So the team changed the frame: agents became digital twins of employees the customer already knew. A worker went home; the twin continued the routine. Resistance eased.
That is not proof that a name makes automation safe. It is evidence that adoption has an interface, and sometimes the interface is a relationship. Cargofy also gives customers control over what an agent can decide alone, what data it can see and where human approval remains mandatory.
The security objection required a more structural answer. Large logistics companies did not want rate history, carrier relationships and contracts leaving their infrastructure. In 2026 Cargofy released ATLAS, an open-source server using the Model Context Protocol. It runs inside a customer's perimeter, indexes logistics data locally and answers an agent's questions without handing the vendor the raw operating archive. The repository is Apache 2.0 licensed and supports road, ocean, air, rail and multimodal records.
The useful part to copy
Cargofy's playbook is more portable than its product. Do the manual work until the exceptions stop surprising you. Pick a narrow pain with a visible meter - fuel, bids, loaded miles, document time. Put the agent on top of the customer's existing tools. Leave judgment and approval with a person. Then widen the job only after the first result earns permission.
The conditions matter
- A customer needs a repeatable workflow, usable data and someone willing to explain the real process.
- The first deployment should have a measurable baseline and a bounded permission set.
- If the process is already efficient, the agent may add ceremony rather than value.
- If leadership expects a vendor to discover, define and own the workflow alone, the launch is likely to stall.
Vozniak says about one in five skeptical prospects believes its small internal R&D team can build the same capability, only to return months later after spending heavily. He is equally frank that Cargofy has had many failed launches. The common problem is not a weak demo. It is a customer who wants an autonomous agent without participating in the definition of autonomy.
The company now names Kaspi, Metinvest and Zammler among customers, says it serves more than 2,000 teams and reports processing over 100,000 loads a day. In June 2026 it raised a $6 million Series A from u.ventures, Toloka.vc and Movens Capital, with Intercom co-founder Des Traynor participating. Axios reported a valuation of about $50 million and a separate $5 million secondary transaction. The primary funding followed a $2 million seed round in 2022.
What Cargofy has built is neither a magical freight brain nor a prettier load board. It is closer to a very fast junior operator with a phone, an inbox, a rulebook and explicit limits. The bet is that logistics does not need another place to look. It needs something willing to make the next call.