Rethinking AI for Mac

Most AI assistants still struggle with fragmented workflows, privacy concerns, and shallow execution. They promise productivity but leave users juggling tabs, apps, and subscriptions. That’s why we built Eney—a Mac-native assistant designed to work intuitively, run locally, and adapt to real user context. Backed by 17 years of macOS expertise and grounded in the latest AI research, Eney isn’t just a tool—it’s a foundation for a more intelligent, integrated way to interact with your Mac.

Read our Vision article here

Delivering on this vision meant rethinking how AI assistants are built from the ground up. Our team developed custom architecture to support Eney’s unique capabilities. Here are the key components.

How It Works: The Technical Pillars of Eney

On-Device & Cloud Inference. Eney follows a hybrid inference model. When you enter a prompt, Eney first classifies its complexity:

  • Simple tasks (e.g. working with calendar, file operations, text transformations) are handled entirely on-device using built-in ML and LLM models for speed and privacy.
  • Complex tasks that involve multi-step logic, reasoning, or external API calls are routed to our secure cloud backend.

Local Engine. At the heart of Eney is a local-first architecture. It runs key workflows directly on your Mac, enabling fast, private, and offline functionality. A special local classifier decides if a user query should be handled by a local or remote engine. Local ones are handled by lightweight, optimized models tailored specifically for macOS. These include:

  • o10r-llama-parameters-extractor for prompt parameter parsing. It’s powered by fine-tuned Llama-3.2-3B-Instruct.
  • o10r-relevance-ranker to prioritize relevant tasks or documents
  • o10r-embeddings-encoder for semantic search and similarity matching

Mnemos. To ground this intelligence in valuable user interaction, we built Mnemos—AI’s context system. Mnemos continuously indexes your macOS environment. Its job is to provide highly personalized and situationally appropriate context. Through Mnemos, Eney continuously collects and maintains contextual data, allowing each new request to benefit from the knowledge of the file system, content of files, and data stored inside closed data formats like Calendar.

  • mnemos-bge-small-en-v1.5 for contextual search in text documents (CoreML Linguistic Tagger, MacPaw)
  • UserInputTagger for parsing user input to extract tags like file type or location (CoreML, MacPaw)
  • mnemos-mobileclip-s2-image for contextual search in image-based content (MobileCLIP, Apple)
  • mnemos-mobileclip-s2-text for contextual search in text embedded in images (MobileCLIP, Apple)

In the future, it will expand to support custom contexts and developer-defined modules.

Building for the Future, Together

Eney is designed to grow beyond our own roadmap, enabling seamless vendor app integrations that bring your product’s capabilities directly into the experience. These contributions unlock exponentially more value for users—far beyond what MacPaw alone could deliver. Two key technologies make this possible:

  1. Vendor Extension Kit (Eney SDK). macOS framework that allows partners to extend Eney with custom skills with advanced UI and domain-specific logic. Our first wave of Setapp Vendors—starting with apps like Permute—is already building with the SDK. Developers interested in joining can sign up via our vendor form.
  2. Expert Zone (AI generated skills). When Eney understands your request but doesn’t have a pre-built skill, it can attempt to generate one using GenAI. Currently in private beta and available via request, this feature paves the way for a more autonomous assistant—one that can code its own tools in real time.

Join the Ecosystem

Eney is more than a product—it’s a platform for a new kind of human-computer interaction: intelligent, local, and deeply collaborative. We’re just getting started. If you’re ready to shape the next generation of software, join us here.