MacPaw and Liquid AI Partner on On-Device Inference for SetApp Developer Ecosystem

On-Device Inference Infrastructure for SetApp

According to TechCrunch, MacPaw, the Ukraine-based developer behind the SetApp subscription app store, has entered a partnership with Liquid AI to build locally hosted inference capabilities. The collaboration centers on Elix, an on-device inference system designed to power MacPaw’s Eney AI assistant and eventually serve as a developer platform. This move positions SetApp—which claims over 150,000 paying users—as an emerging platform for AI-native applications prioritizing local processing and privacy.

MacPaw CEO Oleksandr Kosovan emphasized that locally hosted models will enable users to run AI assistants and agentic workflows offline, reducing dependence on cloud connectivity and external API calls. The partnership addresses a growing market segment where developers and users prioritize on-device execution over cloud inference, a trend accelerated by both privacy concerns and regulatory scrutiny in Europe, where MacPaw operates.

Liquid AI’s Hardware-Optimized Approach

Liquid AI co-founder and CEO Ramin Hasani differentiated the company’s model architecture from existing on-device solutions. According to TechCrunch, Hasani stated that Liquid AI selects hardware-specific architectures before training, yielding more efficient inference directly on device. Beyond architectural optimization, Liquid AI is building a customization stack enabling models to adapt and improve over time using user input—a feature positioning the system as evolving rather than static.

This design philosophy stands in contrast to Apple’s existing local models, which Hasani acknowledged but distinguished on performance grounds. The emphasis on adaptability signals Liquid AI’s belief that on-device models must personalize to user behavior to justify their computational overhead, a capability that cloud-based competitors can more easily provide.

SetApp’s Evolution Toward AI-First App Economics

MacPaw is restructuring SetApp’s monetization around AI workloads. The company is experimenting with credit-based pricing in which users receive an allotment of credits corresponding to AI operations, with credit consumption scaled by task complexity. This model decouples pricing from per-app subscription economics, allowing users to mix multiple AI-powered tools within a single credit pool—a shift toward platform-level consumption accounting.

MacPaw also plans to position SetApp as a hybrid infrastructure platform, offering developers both on-device inference via Elix and cloud model access from providers including Google. This multi-tier approach targets developers who want flexibility to route workloads between local and remote execution depending on latency, privacy, and cost constraints.

Why This Matters

The partnership exemplifies a fragmentation in AI inference infrastructure: cloud platforms like OpenAI and Anthropic dominate the high-capability end, while on-device inference remains fragmented across proprietary systems (Apple), open-weights frameworks (Ollama, llama.cpp), and now specialized platforms like Liquid AI. MacPaw’s bet on Liquid AI suggests that app-store operators see local inference as a differentiator for developer recruitment and user retention—particularly in regions with stricter data-residency regulations.

For developers building on SetApp, the combination of on-device and cloud options reduces lock-in and enables cost arbitrage. For MacPaw, the credit-based pricing model and developer tooling could shift SetApp from a subscription aggregator into an AI-application platform, competing directly with traditional app stores on capability. However, the success of this model depends on whether Elix’s optimization claims hold under production workloads—a test that will require independent reproduction beyond Liquid AI’s own benchmarking.