YesPress / Company File No. 0422092218
Unframe.
The managed AI delivery platform that ships production enterprise AI in days - then bills you for what actually works.
The managed AI delivery platform that ships production enterprise AI in days - then bills you for what actually works.
There used to be a meeting on the calendar at most Fortune 500s called "AI Steering Committee." It convened monthly. It produced slides. It produced more slides. It almost never produced anything that ran in production.
That meeting still exists. It just has fewer attendees now. A growing share of the people who used to sit in it are off doing something else - watching an Unframe-built workflow ingest a thousand commercial real estate leases, or trigger a quote-to-cash sequence at 3 a.m., or quietly catch a fraud signal before the analyst's coffee goes cold.
Unframe is a Cupertino-headquartered managed AI delivery platform that was founded in 2024 and, by May 2026, had booked $100 million in total contract value, raised $100 million in venture capital, and grown to about 130 people split across California, Tel Aviv and Berlin. None of that is the interesting part. The interesting part is how fast its customers are buying the second thing.
By late 2023 the enterprise AI conversation had a familiar shape. Vendors arrived with platforms. Buyers nodded. Pilots launched. Then, almost without exception, the pilots stalled - mired in data cleanup, integration work, governance reviews, hallucination risk, model selection paralysis, or the simple problem that the proof-of-concept ran on a sanitized dataset that bore no resemblance to the actual mess inside the company.
The story repeated itself enough times to become a genre. Analyst reports estimated that the great majority of enterprise AI projects never made it to production. The technology was extraordinary. The delivery model was not.
Shay Levi, Larissa Schneider and Adi Azarya - veterans of the cybersecurity startup Noname Security, which Akamai acquired for roughly half a billion dollars - looked at the gap between AI's promise and the slide decks describing it, and concluded that the missing piece was not another model. It was a delivery layer. Someone, they reasoned, had to be willing to show up, do the integration work, ship the thing, and only get paid when it ran.
The product they built has a name that gives the game away: The Framery. It is, in their language, an OS for production AI - a workshop of pre-configured building blocks that snap together into tailored solutions for a specific operational problem. An agent orchestrator with built-in guardrails and observability. A knowledge fabric that wraps raw enterprise data in the business logic it lacks on its own. A connectivity layer with pre-built integrations for the systems that already run the place. Modular components for search, reasoning, automation, agentic workflows.
The pitch is unusually concrete for an AI company. Unframe does not sell access. It sells outcomes - which, depending on the customer, might be a lease abstraction engine, a portfolio risk assistant, an ops co-pilot for a telecom, or a compliance reporting agent that finally talks to the seven systems that always refused to talk to each other.
It is also unusually patient. The platform is model-agnostic by design. If a customer wants to run on an open-weight model on-prem because their compliance officer says so, Unframe runs on an open-weight model on-prem. If they want a frontier model in the cloud, they get a frontier model in the cloud. The Framery does not have a horse in the model race. It has a horse in the deployment race.
FIG. 1 — The numbers Unframe leads its decks with. None of them are the model parameters.
Strip away the architecture diagrams and what Unframe sells reduces to something concrete. A senior account team listens to a problem. The Framery is configured against that problem - knowledge fabric tuned to the company's data, connectors wired into whatever systems run the workflow, agents orchestrated with guardrails specific to that industry. A working version ships in days. The customer uses it. Outcomes are measured. Pricing follows the outcome.
The catalogue of what enterprises have actually deployed reads less like an AI vendor's wishlist and more like a tour through the unglamorous engine room of a modern company: ticket triage, root cause analysis, lease abstraction, portfolio risk, fraud detection, business intelligence, knowledge extraction from unstructured documents, customer support reasoning, operational reporting, compliance summarization. Useful. Boring. Worth a lot of money to the people who have to do it.
It is also why the expansion data is so unusual. A company that solves one painful workflow tends to discover, immediately afterwards, that it has six more.
FIG. 2 — Cumulative total contract value reaching $100M inside twelve months. Quarterly distribution is illustrative; the cumulative figure is the disclosed one.
The list of named customers is, by enterprise AI standards, unusually un-tech-flavored. Commercial real estate giants Cushman & Wakefield and Avison Young. The Swiss newspaper of record NZZ. Cybersecurity firm Armis. IT services and consulting names Acora, Credera and Climb Global Solutions. Telecom outfit MasterTel. None of these companies are in the business of betting on AI vendors for sport. They are in the business of running operations and either they shipped something that worked, or they did not.
The investor list reads similarly. Bessemer led early. Craft Ventures, TLV Partners, Third Point Ventures, Cerca and Vintage Investment Partners followed. Highland Europe arrived in May 2026 to lead the Series B - the kind of trans-Atlantic vote that does not happen unless the European side of the customer base is already paying.
If you had to compress the company's worldview into one sentence, it would be this: enterprise AI is a delivery problem, and the firm willing to actually deliver wins. There is a polite version of this argument and an impolite one. The polite version goes in board decks. The impolite version is the one that explains why Unframe's customers come back for use case number two, then number three.
The polite version notes that the platform is model-agnostic, cloud-or-on-prem, with pre-built integrations across enterprise systems, and that it ships solutions in days rather than quarters. The impolite version notes that this is what the entire industry promised and very few of the players actually do.
Unframe's leadership talks about being a managed service rather than a tool vendor - the distinction matters. A tool vendor sells you the workshop and wishes you luck. A managed service walks in, builds the thing, runs the thing, and stakes its compensation on whether the thing works. It is an old idea, dressed in new clothes, and it turns out the new clothes fit.
Return to the meeting. The AI Steering Committee. Once a month. Slides. More slides.
It is still on the calendar at many companies. But increasingly, when someone asks for an update, the answer is not "we are in pilot." The answer is "we shipped that one, and the next one, and we are renegotiating the contract because we want more." Whether Unframe specifically becomes the dominant managed AI delivery layer of the 2020s is a question that, like all such questions, will be settled by the customers. The early evidence is encouraging in a way that is hard to fake. Contracts are signed. Revenue compounds. Use cases multiply.
Two years in, Unframe has done something that the broader market spent a decade saying was impossible: it has made enterprise AI feel less like an experiment and more like an installation. That is not nothing. In a category where most companies still ship hope, shipping something that runs in production is, quietly, the move.
The conference room will keep meeting. Just with fewer attendees, and shorter agendas, and - if Unframe's customers keep buying the way they have been - a different kind of question on the table. Not "should we try AI?" but "which workflow next?"
Where to find Unframe, and what to read next.