MacPaw taps Liquid AI to offer on-device inference to devs building for its app store
TechCrunch Ivan Mehta
MacPaw is teaming with Liquid AI to run its Eney assistant fully on your device instead of the cloud. It's also opening the tech to other developers on its SetApp store, credits and all.
MacPaw, the Ukrainian company behind SetApp, wants its AI assistant Eney to work without ever phoning home. To get there, it's turned to Liquid AI, a startup that builds model architectures specifically tuned to run efficiently on local hardware rather than in some distant data center. The two companies are building an on-device inference system called Elix, along with a local memory layer that lets the assistant remember context between sessions without shipping that data anywhere.
Ramin Hasani, Liquid AI's co-founder and CEO, framed the pitch around privacy and efficiency: pick the right architecture for the chip first, then train around it, and you get intelligence that runs locally without the usual compromises. That's a direct answer to a question hovering over the whole industry right now — whether on-device AI can actually match what cloud models deliver. Hasani insists Liquid AI's models are built for performance parity, not just as a stripped-down fallback for offline mode.
There's an obvious complication here: Apple already ships its own on-device models to developers, for free, baked into iOS and macOS. MacPaw's CEO Oleksandr Kosovan isn't pretending that competition doesn't exist, but he's betting that offering agentic workflows and offline assistants through SetApp — which already has more than 150,000 paying subscribers — gives developers a reason to look elsewhere. The plan is to eventually open up this on-device stack so third-party developers building for SetApp can plug into it directly, alongside cloud models from the likes of Google, turning the store into something closer to a one-stop AI shop.
Money is the other half of this story. MacPaw is testing a credit-based system for AI features inside SetApp, where users spend credits based on how complex a given AI task is. It's a pricing model straight out of the API-billing playbook, now being pushed down into a consumer app store. Whether subscribers who are used to flat monthly fees will tolerate metered AI usage is the real experiment MacPaw is running here, arguably a bigger gamble than the on-device architecture itself.
My take
Betting against Apple's free on-device models is a bold move for a mid-sized app store, but MacPaw's real innovation might be the credit-based pricing test — consumers have been trained on flat subscriptions for a decade, and metered AI usage could be a tough sell even if the tech works flawlessly. Local-first AI is the right instinct for privacy, though; the industry has spent too long assuming everything must run through someone else's cloud.
Read more about this at: TechCrunch
