MacPaw is done renting intelligence. In a move that signals a departure from the industry’s obsession with cloud-heavy API calls, the developer is pivoting toward a proprietary on-device AI stack through a strategic partnership with Liquid AI. The goal is Elix—a custom inference system designed to wring every drop of performance out of Apple Silicon. According to Liquid AI CEO Ramin Hasani, the architecture is tailored to the hardware before training even begins. For MacPaw, which manages the Setapp ecosystem and its 150,000+ subscribers, this isn't just a tech upgrade; it’s a fundamental shift from paying OpenAI’s bills to utilizing the NPU cycles already bought and paid for by the end-user.
Sovereignty via On-Device Architecture
The strategic weight of the Elix stack lies in its total decoupling from cloud pipelines. While Apple promotes its own local models, Liquid AI focuses on a customization stack that allows models to evolve based on user input without data ever leaving the machine. MacPaw CEO Oleksandr Kosovan is positioning this as a reliability play: locally hosted models allow assistants and agentic workflows to function entirely offline. This architecture powers the Eney AI assistant, utilizing a local memory system that treats privacy as a competitive advantage rather than a compliance checkbox.
"Before training our models, we select an architecture that is different and tailored to the hardware," explained Ramin Hasani, CEO of Liquid AI, highlighting the performance advantages of their specific approach.
By moving processing to the edge, MacPaw eliminates the latency and cost overhead of external servers. It’s a pragmatic response to a corporate market that is increasingly allergic to cloud-based memory systems and the potential data leaks they represent.
Shifting the Unit Economics of AI Subscriptions
MacPaw is aggressively rewriting the unit economics of the Setapp store to survive the era of expensive compute. The company is rolling out credit-based pricing, where user operations are metered based on complexity. This transition is vital for margin protection; as MacPaw offloads inference to the user's device via Elix, the marginal cost per query effectively hits zero for the developer. Kosovan intends to open this local processing layer to third-party developers on the Setapp platform, essentially attempting to set an infrastructure standard for independent macOS software.
While Setapp will continue to provide access to cloud behemoths like Google for heavy lifting, the push toward on-device inference is a massive optimization play. Instead of burning venture capital on H100 clusters in the cloud, developers can tap into the idle silicon on a user's desk. This creates a sustainable subscription model where high-activity power users no longer cannibalize a developer's profits. The real test for MacPaw will be whether independent developers see enough value in Liquid AI’s customized stack to bypass the free, native tools Apple is already baking into macOS.