MacPaw and Liquid AI Team Up to Bring Offline AI Assistants to Every Mac
News Summary
Ukraine-based software maker MacPaw, the company behind CleanMyMac and the Setapp app subscription platform, has entered a long-term strategic partnership with Liquid AI, the Boston-based startup known for its compact, hardware-efficient foundation models. The deal, announced on August 5, 2026, 9:00 AM Eastern Time, aims to build a shared on-device AI stack for macOS that lets Mac apps run intelligent assistants and agentic workflows entirely offline, without sending user data to the cloud.
What Was Announced
The partnership combines Liquid AI's Liquid Foundation Models (LFMs) with two proprietary technologies MacPaw has been developing in-house: Elix, an on-device inference engine tuned for Apple silicon, and Mnemos, a local memory layer that lets an AI assistant retain context across sessions and grow more useful over time. Liquid AI will tailor future LFM variants specifically for macOS hardware, while MacPaw contributes the inference and memory infrastructure already running in production.
Liquid AI CEO Ramin Hasani said the company's approach starts with hardware, not the model: "Before training our models, we select an architecture that is different and tailored to the hardware it will run on." MacPaw CEO Oleksandr Kosovan framed the payoff for end users as independence from network connectivity, saying local models give "users the ability to run assistants and agentic workflows offline."
Eney Gets the First Local Build
Eney, MacPaw's AI assistant for macOS, is the first product to benefit from the collaboration. Eney had already begun shifting toward local processing, keeping reasoning, contextual search, skill execution, and conversation history on the device whenever possible. The Liquid AI partnership is expected to accelerate that transition, with a fully local build of Eney planned for later in 2026. MacPaw says the goal is an assistant that responds quickly, keeps personal data on the Mac, and continues handling core tasks even without an internet connection.
Opening the Stack to Developers
Beyond Eney, MacPaw plans to extend the shared AI stack to third-party developers through Setapp, its subscription platform that serves more than 150,000 paying users. Once the on-device architecture is finalized with Liquid AI, developers building for Setapp will be able to tap local inference for their own apps rather than building similar infrastructure from scratch. MacPaw also intends to let participating apps expose capabilities and context to Eney, so the assistant can understand what is happening inside other apps and offer more targeted help.
To avoid locking developers into a single model provider, the platform will additionally offer access to cloud-hosted models from companies such as Google, positioning it as a one-stop option for developers who want to mix local and cloud inference depending on the task. MacPaw has indicated that AI operations across Setapp will move toward a credit-based pricing structure as usage grows.
Why On-Device Inference Matters
The push toward local AI processing reflects a broader industry shift as foundation model makers work to shrink models enough to run efficiently on consumer hardware. Liquid AI's recent LFM2 model family illustrates the trend: the company's newest flagship, LFM2-24B-A2B, is built as a mixture-of-experts model designed to move workloads from the cloud to AI PCs and edge devices. In benchmark testing on an Apple M4 Pro chip using INT8 quantization, the model reached roughly 229 tokens per second during prompt processing and about 27 tokens per second during generation, while tests on a Qualcomm Snapdragon 8 Elite chip showed around 35 tokens per second in decoding.
Liquid AI has built out a broader partner ecosystem to support this direction, working with cloud providers including Together AI and Modal, chipmakers such as Intel, AMD, and Qualcomm, and inference tooling projects including Ollama, LM Studio, Cactus, and Nexa AI. AMD's Ramine Roane said the company was "proud to provide day zero support for the latest Liquid Foundation model from Liquid AI," citing its GPU and NPU-enabled processor lineup as a fit for efficient on-device deployment. Cactus co-founder Henry Ndubuaku highlighted coding performance in particular, saying one of Liquid AI's models "excels at coding," and that he was "keen to see on-device coding agents built with these."
What Comes Next
Neither company has disclosed a precise launch date for the developer-facing version of the stack or detailed the underlying model architecture that will ship inside Eney. MacPaw has said results from the partnership are expected later in 2026, starting with the local version of Eney before the broader developer rollout through Setapp. For the millions of Mac users worldwide who rely on privacy-sensitive workflows, the collaboration signals a step toward AI assistants that can operate fully offline while still improving over time through on-device learning.