How Liquid AI Is Breaking the Transformer Monopoly to Power MacPaw's On-Device App Store

MacPaw, the software studio behind the popular macOS curation platform Setapp, is partnering with MIT spinoff Liquid AI to bring on-device inference directly to desktop developers. The collaboration will see MacPaw build a fully local version of its proprietary AI assistant, Eney, while opening up Liquid AI’s highly efficient model architecture to third-party developers building for the Setapp ecosystem. It is a quiet but significant shot across the bow for cloud-dependent AI providers, signaling that the future of desktop software belongs to highly optimized, local execution.

The Architecture Shift: Liquid Neural Networks vs. The Transformer

For the past three years, the tech industry has behaved as if generative AI and the transformer architecture are synonymous. However, as developers try to squeeze large language models onto consumer hardware, the fundamental limitations of transformers—namely their quadratic computational complexity and massive memory footprint—have become glaring bottlenecks. For a desktop application running in the background of a user’s MacBook, keeping a massive transformer model warm in unified memory is a non-starter.

This is where Liquid AI, founded by a team of prominent researchers from the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), enters the picture. Liquid AI pioneered Liquid Neural Networks (LNNs), a class of bio-inspired AI models that utilize continuous-time differential equations instead of static weight matrices. Unlike traditional transformers, which scale quadratically ($O(N^2)$) with context length, Liquid AI’s architecture features a constant state size, allowing it to process sequential data with unprecedented computational and memory efficiency.

By deploying Liquid AI’s models locally, MacPaw is bypassing the resource-heavy overhead of standard transformer models. The partnership is a major validation of alternative AI architectures, proving that non-transformer models are not just academic experiments but viable, commercial-grade engines capable of running on consumer-grade Apple Silicon.

Empowering the Setapp Developer Ecosystem

While MacPaw is using the technology to rebuild its digital assistant, Eney, the broader implications lie in how this technology will be distributed. MacPaw plans to expose Liquid AI’s local inference pipeline as an API for developers publishing software on Setapp, its curated app subscription service.

"By providing developers with highly efficient, on-device AI capabilities, we are eliminating the massive API cost barriers that have prevented indie developers from fully adopting generative features."

MacPaw Product Strategy Team

Currently, desktop utility developers face a brutal economic equation when integrating AI. If they rely on proprietary cloud APIs like OpenAI's GPT-4o or Anthropic's Claude, they must absorb variable, recurring API costs while charging users a flat monthly subscription. This margin compression is existential for small indie developers. By leveraging Liquid AI’s local models, Setapp developers can offload the compute cost entirely to the user's on-device Neural Engine, shifting the marginal cost of AI inference to zero.

Why Local Inference Wins the UX Battle

Beyond the economics, local inference completely transforms the user experience for desktop applications. When an AI assistant like Eney runs on-device, it gains three distinct advantages over cloud-based counterparts:

• Zero Latency: Without the network round-trip to a data center, interactions feel instantaneous, enabling seamless auto-complete and real-time system actions.
• Absolute Privacy: Sensitive user files, system logs, and codebases do not need to be transmitted to third-party servers, a crucial selling point for enterprise users.
• Offline Availability: Core productivity tools continue to function flawlessly in low-connectivity environments, such as during flights or in remote locations.

As Apple continues to aggressively upgrade the neural processing capabilities of its M-series chips, the hardware barrier to local execution is rapidly dissolving. The bottleneck has shifted from hardware capacity to software efficiency, and Liquid AI's continuous-time models are uniquely positioned to exploit this window.

The Strategic Playbook for Indie AI Development

The MacPaw-Liquid AI alliance represents a preview of the next phase of the AI gold rush. The first phase was characterized by centralized, massive scale, where scale-at-all-costs cloud models dominated. The second phase, now underway, is characterized by hyper-optimization and architectural diversification at the edge.

For founders and developers, the lesson is clear: relying solely on wrapper-style integrations of centralized APIs is a temporary strategy. To build defensible, high-margin software in the AI era, builders must master the art of local deployment. By democratizing access to LNN-powered inference, MacPaw is turning Setapp into a premium incubator for local-first AI applications.

Takeaway

The monopoly of the transformer is fracturing at the edge. To win the consumer desktop, AI must be fast, private, and free of API tax—milestones that are only achievable by matching next-generation neural architectures with local hardware.