MacPaw and Liquid AI Partner to Build a Private On-Device AI Stack for macOS

• Liquid Foundation Models (LFMs) – Liquid AI is designing and fine-tuning specialized LFMs specifically for macOS AI assistance. These models are architected to be highly efficient on Apple Silicon .
• Elix – MacPaw's local inference engine that runs the LFMs entirely on the Mac. It enables the models to execute on-device without sending data to the cloud, giving users fast responses, privacy, and offline capability .
• Mnemos – MacPaw's persistent memory layer that retains context across interactions, so the assistant "remembers" previous conversations and becomes more useful over time .

Eney: The First Product on the Stack

Eney is MacPaw's AI assistant for macOS. It will be the first product powered by the new stack, delivering on-device reasoning, contextual search, skill execution, and conversation history—all kept local. Cloud models will remain available as an option when a task is better suited to the cloud . Eney already shifted to a local model last year, and the new partnership is intended to deepen that local architecture .

Developer Access via Setapp with Credit-Based Pricing

Once the stack is proven in Eney, MacPaw plans to open the same on-device inference technologies (Elix + Mnemos + LFMs) to third-party developers through Setapp, its subscription-based app marketplace with over 150,000 paying subscribers .

• Credit-based pricing – MacPaw is experimenting with a system where Setapp users purchase credits. Each AI operation (local or cloud) consumes credits based on the complexity of the task .
• One-stop shop for developers – Setapp developers will be able to offer on-device AI features via the shared stack, plus access cloud models from providers like Google if a task benefits from cloud-scale compute . This means developers get a single platform with both local and cloud inference options.

The partnership is framed as a privacy-first alternative to cloud AI, with the goal of keeping personal data on-device by default while letting users reach cloud models when they are the better tool for a given task .