MacPaw teams with Liquid AI to bring on-device AI to SetApp and developers

MacPaw is working with Liquid AI to build locally hosted AI models for its products and, later, to offer that same on-device inference technology to outside developers. The move matters because it could let apps run assistants and agentic workflows with more privacy, lower latency and offline capability, while also giving MacPaw a new platform play around its SetApp app store.

The Ukraine-based software maker, best known for Mac utilities and subscription services, said the partnership will help power a locally hosted version of its AI assistant Eney and underpin a new local memory system and inference stack called Elix. MacPaw is also preparing SetApp for a broader wave of AI apps, with a credit-based pricing model designed to meter usage by task complexity.

MacPaw’s next AI bet is local, not cloud-first

MacPaw has spent the past year laying groundwork for an AI assistant product called Eney. Now the company is pushing that effort further by building a version that can run directly on a device rather than relying entirely on remote servers.

That shift is central to the company’s partnership with Liquid AI, a startup focused on small, efficient models that can be tailored for different devices. Liquid AI will help MacPaw develop Elix, its on-device inference system, along with a local memory layer intended to support more personalized assistant behavior without sending everything to the cloud.

In practice, that means some AI tasks could happen locally on a user’s laptop, improving responsiveness and potentially reducing dependency on constant internet access. MacPaw says the approach should also support assistant features and agent-like workflows while users are offline.

Why does on-device inference matter?

On-device inference matters because it can keep sensitive data on the user’s machine, shorten response times and allow certain AI features to work without a network connection. For consumer software companies, those benefits are increasingly important as users ask more questions about privacy, reliability and cost.

Ramin Hasani, Liquid AI’s co-founder and chief executive, said the company begins by selecting model architectures designed for specific hardware, which allows it to deliver intelligence that runs directly on devices with stronger privacy and security characteristics.

MacPaw chief executive Oleksandr Kosovan said local models would also make it possible for users to run assistants and agentic workflows offline, broadening the range of circumstances in which AI tools remain useful.

How is Liquid AI different from conventional model providers?

Liquid AI is positioning itself around hardware-aware model design rather than the bigger-is-better logic that has defined much of the AI race. Instead of adapting general-purpose models after the fact, the company says it starts by choosing an architecture suited to the target device.

That approach is meant to improve efficiency and bring performance gains for specific use cases. Hasani said Liquid AI is also building a customization stack that allows models to adapt using user input over time, making them more capable as they interact with more data.

For MacPaw, the appeal appears to be a mix of technical control and product differentiation. Rather than relying only on external model APIs, the company wants a system tuned for its own assistant and, eventually, made available to developers building on its platform.

What is Elix?

Elix is MacPaw’s name for the on-device inference system being developed with Liquid AI. It is intended to serve as the local processing backbone for Eney and, later, potentially for apps built by third-party developers using MacPaw’s ecosystem.

While MacPaw has not publicly detailed every technical component, the company describes Elix as part of a broader local AI stack that includes memory handling and model execution on the device itself. That architecture is meant to support better privacy and lower latency than a cloud-only model path.

SetApp is becoming an AI distribution channel

MacPaw’s subscription app store, SetApp, is becoming a more important part of the company’s AI strategy. The service already has more than 150,000 paying users, giving MacPaw a meaningful base from which to introduce AI-powered tools and usage-based billing.

The company plans to make SetApp more AI-focused over time, with locally run apps and cloud-connected services sharing space in what it hopes will be a one-stop environment for developers and customers. Once the local architecture is finalized, MacPaw wants to expose it to third-party developers so they can use on-device inference inside their own applications.

That could make SetApp not just a storefront, but also a distribution and infrastructure layer for AI software. MacPaw says developers would be able to access local processing tools through the platform, alongside cloud models from providers such as Google.

How will pricing work inside SetApp?

How pricing will work is still being tested, but MacPaw is experimenting with a credit-based model that charges users according to the number and complexity of AI operations they perform.

That is a notable departure from flat subscription pricing because AI tasks can vary widely in compute cost. A short text rewrite may consume far fewer resources than a complex workflow that chains multiple model calls, memory lookups and agentic steps.

For users, the system could make spending more predictable and more closely tied to actual usage. For MacPaw, it provides a mechanism to monetize AI features without forcing every customer into the same consumption pattern.

Apple is already local-first, so what’s MacPaw’s angle?

Apple already offers developers access to some of its own local AI capabilities, which means MacPaw is entering a field with established competition on the same platform. But MacPaw’s pitch is not that it is inventing local AI from scratch; it is trying to assemble a broader, developer-friendly stack that can combine local inference, memory, and optional cloud models.

That bundling could matter for independent developers who want flexibility rather than a single vendor approach. If MacPaw can offer one workflow for local execution and another for cloud fallback, it may appeal to app makers who want to balance privacy, performance and feature depth.

The company’s larger opportunity may be in software distribution. By turning SetApp into an AI-ready ecosystem, MacPaw can potentially position itself as both a tooling provider and a channel for discovering and paying for AI applications.

What MacPaw and Liquid AI are building

The partnership spans several layers of the AI stack, from model selection to execution and memory. The table below summarizes the major pieces MacPaw has discussed so far.

Timeline: how MacPaw’s AI strategy is evolving

The company’s AI plans have progressed in stages, moving from an assistant concept to infrastructure and then to a possible platform play for developers.

What this means for developers

For developers, MacPaw’s plan could offer a rare combination: local AI execution for privacy-sensitive features, plus optional access to cloud models when heavier lifting is needed. That hybrid model could reduce integration complexity for smaller teams that do not want to build an AI stack from zero.

It could also lower friction for apps that need to operate offline or in constrained environments. A note-taking app, productivity tool or system utility may benefit from local summarization or classification without transmitting user data to a server on every request.

MacPaw’s bet is that developers will value an environment where they can choose the right model path for each use case. If the company succeeds, SetApp could become an AI infrastructure layer as well as a marketplace.

Why this partnership matters now

The timing reflects a broader industry shift toward smaller, more efficient models and more practical deployment options. As the costs and limitations of large cloud-first systems become clearer, more companies are exploring whether some AI workloads can move onto user devices.

That trend is especially relevant for companies building productivity software, where speed, discretion and offline availability can be major selling points. It is also relevant for app stores and software marketplaces, which are looking for new reasons for users to subscribe and for developers to distribute through them.

MacPaw’s move suggests it sees AI not only as a feature layer, but as a business model and platform opportunity. By pairing Eney with SetApp and a local inference stack, the company is trying to control more of the experience from model execution to distribution.

Key questions around the MacPaw-Liquid AI deal

Will users notice the difference?

Yes, if the system works as intended, users may notice faster responses, better privacy and more resilient offline functionality. The effect would be most visible in tasks that can be handled locally without waiting on a remote server.

Will cloud AI disappear from SetApp?

No, MacPaw says cloud models will still be part of the picture. The company wants to offer both local inference and access to external models, including providers like Google, rather than forcing one architecture across every task.

Is this only about MacPaw’s own products?

No, the longer-term plan is to make the technology available to other developers. That could widen the market for the local inference stack beyond Eney and into third-party apps distributed through SetApp.

The bigger competitive picture

MacPaw is entering a fast-moving area where platform companies, AI labs and startup infrastructure providers are all trying to define the rules of application-level AI. Apple’s local model offering means device-level AI is already part of the operating-system conversation, while cloud providers continue to push their own APIs for app developers.

MacPaw’s advantage may lie in focus. Rather than trying to be everything at once, it is targeting a specific user base, a known distribution channel and a clearly defined software category: productivity and utility apps that benefit from privacy and offline access.

If the company can prove that local AI is practical inside mainstream apps, it may be able to persuade more developers to build with a hybrid model in mind. That would give SetApp a stronger identity at a time when software marketplaces are under pressure to offer more than just downloads.

For now, the announcement is less about a finished product than a direction of travel. MacPaw is signaling that the next phase of its AI work will be device-first, developer-facing and tied closely to the economics of software distribution.

That combination could make the company one to watch as the market moves from chatbot demos toward real deployment choices about where AI should run, how it should be billed and who controls the stack.

MacPaw’s chief executive said the company wants SetApp to evolve into a place where developers can access both local inference and cloud AI, making it easier to build and distribute AI-enabled applications in one environment.

The strategy is ambitious, but it fits the direction the industry is heading. As AI becomes less about novelty and more about infrastructure, the companies that can make models efficient, private and easy to deploy may gain an edge over those still selling raw access to increasingly crowded cloud APIs.