MacPaw Brings On-Device AI Inference to Its App Store via Liquid AI
MacPaw is partnering with Liquid AI to power on-device inference for its AI assistant Eney, with plans to expose that capability to developers building for MacPaw's Setapp app store. The move positions MacPaw as an on-device AI distribution layer for Mac-focused developers.
Original sourceMacPaw, the Ukrainian software company best known for CleanMyMac and the Setapp subscription app store, is integrating Liquid AI's small language models to run its AI assistant Eney locally on device. The partnership extends beyond MacPaw's own products: the company plans to offer the on-device inference capability as a service to third-party developers who distribute through Setapp, giving them a way to ship AI features without routing user data to the cloud.
Liquid AI, the MIT-spinout known for its Liquid Foundation Models built on an alternative architecture to traditional transformers, is a notable choice here. The company has focused on efficient inference at smaller parameter counts — a meaningful fit for on-device constraints where memory bandwidth and thermal limits matter more than raw GPU throughput. MacPaw hasn't disclosed which specific model sizes are being deployed or what the hardware floor is for end users.
The strategic logic for MacPaw is clear: Setapp competes with the Mac App Store partly on developer experience, and bundling accessible on-device AI inference could be a real differentiator for indie developers who can't afford to build and maintain cloud AI infrastructure. For users, on-device inference means privacy by default — a meaningful pitch in a post-GDPR, post-ChatGPT-data-scandal world where local processing is increasingly a selling point rather than a constraint.
Details on the developer-facing API, model access tiers, and whether this is Setapp-exclusive or available more broadly remain thin. MacPaw has not published documentation or a developer preview timeline, which makes it difficult to evaluate the implementation beyond the partnership announcement itself.
Panel Takes
The Builder
Developer Perspective
“The primitive here is a managed on-device inference runtime exposed to third-party developers — potentially useful if the API surface is clean and the model quantization story is transparent. But right now there's no repo, no docs, and no published hardware floor, which means this is a partnership press release, not a developer tool. Ship when there's an SDK with a working hello-world that doesn't require five configuration steps to tell me which Liquid model variant I'm actually running.”
The Skeptic
Reality Check
“The category is on-device inference SDKs for desktop, and the real competition is Apple Intelligence's on-device stack — which ships free with macOS and doesn't require a third-party app store relationship. MacPaw's moat here is Setapp distribution, not the AI capability itself, and that's a thin moat if Apple decides to open its on-device models to any App Store developer. What kills this in 12 months: Apple ships broader on-device model access in macOS 27 and the differentiation collapses entirely.”
The Futurist
Big Picture
“The thesis MacPaw is betting on: app stores that bundle AI inference will replace app stores that don't, because indie developers won't maintain cloud AI infrastructure themselves. That's a plausible claim — the cost and compliance burden of running inference in the cloud is real friction for solo developers. The second-order effect worth watching is whether this makes Setapp a genuine AI distribution layer rather than just a software subscription bundle, shifting power away from per-app AI feature races toward whoever controls the runtime.”
The Founder
Business & Market
“The buyer here is the indie Mac developer who's already in Setapp and needs AI features without the overhead of standing up cloud infrastructure — that's a real person with a real problem. The moat question is whether MacPaw can make the inference layer sticky enough that developers build workflows around it, because if Liquid AI decides to go direct-to-developer, MacPaw is just a reseller. The business only works if the distribution relationship with Setapp's developer base creates lock-in that the model provider can't easily bypass.”