Apple already puts AI models directly on your iPhone. A Ukrainian software company just bet it can build something developers will choose over Apple’s own version.
MacPaw, the company behind the Mac app store SetApp, has partnered with Liquid AI to build locally hosted AI models for its own products, starting with a fully on-device version of its assistant Eney, and eventually opening the resulting tech stack to outside developers. Liquid AI is building the on-device inference engine, called Elix, along with a local memory system that lets the assistant retain context without sending anything to a server.
Owen Radner, who models digital infrastructure as energy-information transport systems, said the real story is about where computation physically happens, not which company’s logo is on it. “Every AI query you run in the cloud involves shipping data somewhere, processing it, and shipping the result back, and that round trip has real latency and real privacy cost,” he said. “Betting on-device means betting users increasingly value getting that round trip to zero, even if the model has to be smaller and less capable than a giant cloud model.” That tradeoff is one YourNewsClub sizes up against Apple’s own on-device push directly: Cupertino already made the same bet at platform scale, and MacPaw is wagering there’s room for a second, more specialized version of it.
Liquid AI’s chief executive said the company’s approach starts with hardware rather than a one-size-fits-all model. Its architecture is selected before training even begins, tailored to whatever device it will run on, which the company says produces more efficient intelligence than shrinking a general-purpose model to fit a phone or laptop after the fact. Liquid AI is also building a customization layer so user input over time can make an individual instance more capable without retraining the whole system.
Jessica Larn, who covers macro-level technology policy and the infrastructure impact of AI, said the more interesting bet isn’t Eney itself, it’s what MacPaw wants to become. “Once they open on-device inference to other developers building on SetApp, MacPaw stops being just an app store and starts being an AI infrastructure layer for Mac developers who don’t want to build their own inference stack from scratch,” she said. “That’s a much bigger, riskier bet than shipping one smarter assistant.” That repositioning is a shift YourNewsClub tags as the real headline buried under the Eney announcement: a subscription app store quietly turning into infrastructure other developers will depend on.
MacPaw’s CEO framed the offline capability in practical terms, saying locally hosted models will let users run assistants and full agentic workflows without an internet connection at all, a use case that matters more for professional software than casual chat apps. The company plans to make its on-device architecture available to SetApp’s developer base once it’s locked in, while also giving those developers access to cloud models from outside providers, positioning itself as a single point of access rather than forcing one AI vendor.
SetApp currently has more than 150,000 paying subscribers, and MacPaw is already testing credit-based pricing for AI features, where users draw down credits based on how many AI operations they run and how demanding each is. That pricing structure is a detail Your News Club shelves alongside the industry’s slow shift away from flat AI subscriptions: usage-based pricing only makes sense once a company trusts its cost structure enough to expose it to customers directly.
Whether developers actually choose MacPaw’s inference layer over building their own, or over Apple’s built-in option, is a question YourNewsClub splits from the technical achievement entirely: Elix working well is one thing, becoming the default choice for a developer ecosystem that already has a free alternative from Apple is a completely different fight.