MacPaw taps Liquid AI to run assistants offline, bypassing Apple's models

Ukraine-based MacPaw isn’t waiting around for Apple’s on-device intelligence to mature. The company behind CleanMyMac and the SetApp subscription store has partnered with MIT spin-off Liquid AI to build a locally hosted inference system called Elix, betting that a specialized architecture will deliver better performance and privacy than the generic local models Apple already provides.

Liquid AI’s approach is fundamentally different. Rather than forcing a standard model onto hardware, co-founder and CEO Ramin Hasani told TechCrunch they “select an architecture that is different and tailored to the hardware” to squeeze out maximum efficiency. This isn’t just about slapping a small language model onto an M-series chip. The collaboration also includes a local memory system, and a customization stack that lets models improve using on-device user data. MacPaw CEO Oleksandr Kosovan is selling the offline angle hard—users will be able to run assistants and agentic workflows without an internet connection.

The real play here extends far beyond MacPaw’s own AI assistant, Eney. Once the local processing architecture is locked in, the company plans to open it up as a developer platform. Kosovan envisions a one-stop shop where devs can access on-device inference alongside cloud models from Google and others. Think of it as an abstraction layer that handles the messy plumbing of hybrid AI.

SetApp’s 150,000 paying subscribers are the obvious beachhead. MacPaw is already experimenting with credit-based pricing for AI operations, where the cost scales with task complexity. That’s a pragmatic hedge against unpredictable inference costs, and it suggests MacPaw sees AI features not as a freebie but as a premium tier that could reshape its subscription revenue. The open question is whether developers will trust a third-party stack when Apple inevitably tightens its own on-device AI integration in the next iteration of macOS.