MacPaw is making a smart, slightly ambitious bet: if AI is going to live on the Mac, it should actually live on the Mac. By partnering with Liquid AI, the company is rebuilding its assistant Eney around local inference first, then planning to expose that stack to developers through Setapp.
<> That’s the real story here: not just a better assistant, but a potential local-AI platform./>
The first layer of the plan is Eney, MacPaw’s macOS assistant, which will use Liquid AI’s models alongside MacPaw’s own Elix inference engine and Mnemos memory system. MacPaw says the goal is to run core tasks on-device, with cloud models still available when they are the better tool. That hybrid stance matters. It signals that MacPaw understands the obvious limitation of on-device AI: not every task belongs on a laptop CPU, no matter how elegantly packaged the pitch sounds.
What makes this announcement interesting is not the usual privacy-and-speed boilerplate, though those benefits are real. On-device inference reduces latency, avoids network dependence, and keeps more data on the user’s machine. The more consequential angle is strategic: MacPaw is trying to turn its AI plumbing into reusable infrastructure. If it works, developers could get a shortcut to local AI without having to assemble their own model-serving stack.
That is a meaningful shift for a company best known for utilities like CleanMyMac. MacPaw is no longer just shipping apps; it is trying to become part of the substrate that other Mac apps run on. In other words, this is a product announcement and a platform play wrapped into one.
A few details make the ambition clearer:
- Eney is the proving ground for the architecture.
- Elix handles on-device inference, while Mnemos handles memory.
- MacPaw says the system is designed for Apple silicon and local optimization.
- The company wants to extend the stack to more products and then to developers via Setapp.
<> If MacPaw opens this up cleanly, Setapp could become more than an app marketplace — it could become a distribution layer for local AI./>
There’s also a quiet validation story for Liquid AI. Its device-native models are getting a high-visibility deployment case, and that matters in a market that is still sorting out whether “AI on the edge” is a real category or just a demo-friendly slogan. MacPaw’s move suggests there is genuine demand for fast, private, offline-capable assistants, especially on the Mac where hardware and software are tightly coupled.
Still, the fine print matters. On-device AI is only as good as the device it runs on, which means capability is bounded by compute and memory. That’s why MacPaw’s hybrid design is pragmatic rather than ideological. The company is not pretending local models can do everything; it is saying they should do the things they can do best, and hand off the rest.
My take: this is exactly the kind of AI architecture Mac developers have been waiting for. Not because it is flashy, but because it is boring in the right way — local, fast, privacy-preserving, and integrated enough to feel native. If MacPaw can make that easy to adopt inside Setapp, it may end up selling something more valuable than a feature: a default.
