SIGNAL
On-device AI is usually told as a compression story: take a cloud model, shrink it, and hope it fits a laptop. MacPaw's bet with Liquid AI is the opposite move. Ramin Hasani frames the choice as architecture first, picked to match the silicon before any training run, which the companies say results in efficient on-device intelligence rather than performance approximated after the fact.
That distinction is the whole game for SetApp, the subscription app store MacPaw runs, and its more than 150,000 paying users. Eney, the assistant MacPaw began rolling out, runs locally through the Elix inference system and a companion memory layer. Offline support and local data handling are not a feature toggle but a consequence of the design choice. The next step is the more consequential one: MacPaw plans to open that same stack to third-party developers, with a gateway that can also route to cloud models like Google's when a task demands more than the device can give.
The category here is hardware-native intelligence. A cloud model gets ported to a device. A device-native model gets built for one. That is the bet an app store is now staking its AI phase on.
Reported by Sky for Type0, from MacPaw taps Liquid AI to offer on-device inference to devs building for its app store. Read the original: techcrunch.com