# MacPaw Research > MacPaw Research — advancing Mac-native intelligence. On-device AI, LLM efficiency, AI memory, and human–computer interaction. Papers, tech notes, and highlights from a research department at MacPaw. MacPaw Research is the in-house research department at MacPaw, focused on on-device intelligence for the Mac. We publish peer-reviewed papers, tech notes, and engineering highlights as we go. Topics span on-device LLM inference, model efficiency, AI memory and personalization, multimodal human–computer interaction, security and software analysis, and applied ML for desktop software. This file follows the [llms.txt](https://llmstxt.org/) convention: it gives AI agents and LLMs a curated map of the most relevant URLs on this site. The full content of every article is available as one document at [/llms-full.txt](https://research.macpaw.com/llms-full.txt). Machine-readable feeds: [RSS](https://research.macpaw.com/feed.xml), [Sitemap](https://research.macpaw.com/sitemap.xml). ## Publications Sorted newest first. - [Memory Belongs to the User: Portable Memory, an Open Standard Proposal for Cross-Vendor AI Memory](https://research.macpaw.com/publications/portable-memory): Portable Memory is an open, vendor-neutral format for carrying AI memory across apps, devices, and vendors — lossless, local-first, with verifiable deletion as its integrity guarantee. - [MacArena: Benchmarking Computer Use Agents on an Online macOS Environment](https://research.macpaw.com/publications/macarena): MacArena introduces a comprehensive benchmark for evaluating computer use agents on an online macOS environment, featuring 421 manually verified tasks across 50 applications. - [Help, Don’t Control: Measuring User Emotional Background in AI Assistant Interaction](https://research.macpaw.com/publications/help-dont-control-measuring-user-emotional-background-in-ai-assistant-interaction): An exploratory HCI study using EEG and eye-tracking to examine how AI assistant design shapes users' emotional experience during task execution, identifying pacing, framing, and perceived control as key factors for emotionally aware interface design. - [Composable AI Pipelines: Write Intent, Not Infrastructure](https://research.macpaw.com/publications/composable-ai-pipelines): MacPaw's Composable Pipelines separates AI workflow intent from infrastructure using a SwiftUI-inspired Swift DSL. Developers declare pipeline logic with property wrappers and result builders; the runtime handles dependency analysis, parallel batching, model ranking, and re-execution — keeping product code stable as the execution layer evolves. - [GUIrilla: A Scalable Framework for Automated Desktop UI Exploration](https://research.macpaw.com/publications/guirillascalableframeworkautomated): GUIrilla introduces a scalable method for autonomous macOS UI exploration using accessibility APIs, producing structured GUI graphs, 27k+ synthetic tasks, and a high-coverage dataset for training and benchmarking desktop-level AI agents. - [Fast-dLLM on MLX: Training-Free Acceleration for Diffusion Language Models on Apple Silicon](https://research.macpaw.com/publications/fast-dllm-mlx): An MLX port of Fast-dLLM for Dream-style diffusion language models on Apple Silicon, with dual-cache decoding and parallel token finalization. - [Do Our Design Principles Ever Really Change, or Do We Just Change the Methods of Achieving Them?](https://research.macpaw.com/publications/design-principles-change): The paper explores how foundational interface design principles translate from classic direct manipulation systems to modern AI-driven Infinite Interfaces. - [GLiNER2 for Swift: Unified Schema-Based Information Extraction on Apple Silicon](https://research.macpaw.com/publications/gliner2-swift): GLiNER2Swift brings schema-based information extraction to Swift with Apple Silicon support, enabling on-device NER and classification without Python runtimes. - [Demonstrating Headphone-Sensed, Accessibility-Informed Head Pointing with Snapping for Inclusive Interaction](https://research.macpaw.com/publications/head-pointing-with-snapping): This paper demonstrates an inclusive interaction technique that combines headphone-sensed head pointing with accessibility-informed snapping to enable precise, hands-free user input across diverse accessibility needs. - [Deep Language Geometry: Constructing a Metric Space from LLM Weights](https://research.macpaw.com/publications/digital_language_geometry): Deep Language Geometry uses LLM weight activations to map 106 languages into a metric space, revealing linguistic families and hidden connections. - [Screen2AX: Vision-Based Approach for Automatic macOS Accessibility Generation](https://research.macpaw.com/publications/screen2axvisionbasedapproachautomatic): Screen2AX is a vision-based framework that reconstructs macOS accessibility trees from screenshots, generating real-time structured UI metadata to enhance grounding and screen-reader support. - [Measure Twice, Cut Once: the Best* Way to Calculate macOS File Hierarchy Size](https://research.macpaw.com/publications/disco): Optimizing APFS file system scanning with faster algorithms, multithreading, and traversal methods for efficient storage management. - [Mousetraps: Reliably distinguishing free apps and trial versions of paid macOS apps on public app marketplaces](https://research.macpaw.com/publications/mousetraps): Easily spot free vs. paid macOS apps! See how machine learning improves app labels, helping you make smarter, trust-worthy choices. - [SwiftEval: Developing a Language-Specific Benchmark for LLM-generated Code Evaluation](https://research.macpaw.com/publications/swift_eval): Paper from MacPaw Research. - [Towards Generating App Feature Descriptions Automatically with LLMs: the Setapp Case Study](https://research.macpaw.com/publications/theseus): Paper from MacPaw Research. - [Case Study: Using EEG to Assess Cognitive Load in Infinite Interfaces](https://research.macpaw.com/publications/using-eeg-to-assess-cognitive-load): This research compares cognitive load in Infinite Interfaces versus traditional interfaces to identify any similarities or differences in the results. - [Datasets: Inside Mac App Store Applications Metadata](https://research.macpaw.com/publications/macappsstore-metadata-dataset): Metadata for over 87000 macOS apps, sourced from the public API for the US, Germany, and Ukraine. It contains details like app names, descriptions, and more. - [R+R: State of the Application Sandboxing on macOS: A Differentiated Replication](https://research.macpaw.com/publications/sandbox): Our work confirms the applicability of the original sandboxing analyses in different contexts, such as dataset or test OS environment changes. Let’s dive in! - [Evaluating a multi-method approach for user emotional dynamics assessment in Software UX](https://research.macpaw.com/publications/emotional-assessments-ux): In our study, we combine self-reporting, behavioral observation analysis, direct user speech, and AI-powered facial expression analysis. Let’s take a look! - [Position Paper: Think Globally, React Locally — Bringing Real-time Reference-based Website Phishing Detection on macOS](https://research.macpaw.com/publications/website-phishing-detection-macos): Read more on the potential of on-device, real-time phishing detection systems to enhance cybersecurity defensive technologies. - [Parsing macOS Application UI: Techniques and Tools for Effective Analysis and Automation](https://research.macpaw.com/publications/how-to-parse-macos-app-ui): In the research, we explore how to parse macOS application UI using code tools and compare the advantages and disadvantages of different methods. - [AI Opportunities in Operating Systems](https://research.macpaw.com/publications/ai-in-os): Explore the concept of AI operating systems (AIOS), examining their potential to enhance OS performance, optimize processes, and evolve with machine learning. - [Collecting a Dataset of macOS Apps: Pains, Gains, Lessons Learned. Part 1](https://research.macpaw.com/publications/macos-app-dataset): Our method of collecting macOS apps does not suffer from any regional restrictions imposed by the app stores and results in 12k items. Check out more details! - [Application Streaming: Use macOS Apps Remotely](https://research.macpaw.com/publications/app-streaming): Check out the story of our first attempt at implementing an application streaming prototype. Let’s dive in to learn the current achievements! - [Infinite Interfaces](https://research.macpaw.com/publications/infinite-interfaces): Explore everything about infinite interfaces — design principles, testing methods, potential future directions, and much more. - [Improving macOS Accessibility Features with W1 Chip sensors](https://research.macpaw.com/publications/headphones-accessibility): In this article, we'll cover some ways to improve accessibility using Apple AirPods or any other device with a W1 chip onboard. - [Passwordless Authentication: Possible Implementation of User Authentication During Registration and Login Using Passkeys](https://research.macpaw.com/publications/passwordless-authentication): Passkeys are a reliable and convenient solution to get rid of passwords. Learn more about registering, removing, and sharing passkeys. - [Diskoverse. Visualizing macOS Filesystem Using Augmented Reality](https://research.macpaw.com/publications/diskoverse): Explore Diskoverse — a project visualizing the macOS filesystem using augmented reality to enhance user interaction and system understanding. - [Static Analysis of macOS Application Bundles for Territorial Affiliation](https://research.macpaw.com/publications/static-analysis-country): Static analysis of macOS app bundles is a straightforward and powerful approach for getting the country affiliation of the software. Learn more in our publication. ## Topics - **artificial-intelligence** — 12 publications - **human-computer-interaction** — 8 publications - **security** — 3 publications - **software-analysis** — 7 publications - **software-distribution** — 2 publications - **software-engineering** — 3 publications ## Other - [Partnership](https://research.macpaw.com/#partnership): How we collaborate with universities, research labs, and industry teams. Email contact for research partnerships. - [Full content (single document)](https://research.macpaw.com/llms-full.txt): Concatenated markdown of every published paper, tech note, and highlight on the site. - [RSS feed](https://research.macpaw.com/feed.xml): Subscribe to new publications. - [Sitemap](https://research.macpaw.com/sitemap.xml): Canonical URL index.