In memory of
Catharina
What is vAI?
vAI is a novel and innovative approach to Machine Learning and Artificial Intelligence. Drawing from years of experience and collaborating with natural language processing experts, our clients, and the vaic.at team’s requirements, we have meticulously designed three distinct methods to engage with the next-generation ML/AI implementation, aptly named vAI. These methods encompass our expertise in precise knowledge engineering, neural networks, and automatic technology.
CathAI – Our Ambient AI experience
We perceive a significant opportunity for an ‘Ambient AI’ platform. CathAI, therefore, stands as our most advanced ambient AI platform, intricately connected to the immediate environment it operates within.
CathAI frequently appears at the most convenient moment for the user in any service, app, or experience we develop.
CathAI’s primary functions include connecting and assisting with data parsing for Apple Foundation Models, Apple Intelligence, and vAI. For instance, in the mEUvy app, CathAI can remind users of crucial data from previous searches or comparisons of cities and countries. Additionally, CathAI highlights relevant data and interests based on the user’s profile, all while performing on-device inference.
CathAI released in mEUvy app and will soon be available in the upcoming EUnify app. Implementations for EUorigin, CUEB.ES and Salary Insights are planned for 2026.
CatharinAI – Our most powerful AI expression
CatharinAI, a standalone natural language processing app powered by MLX and proprietary data sets from VIT, mEUvy, EUnify, EUorigin, and CUEB.ES, is our on-device AI companion designed with privacy, speed, and control in mind. Unlike traditional AI solutions that send your thoughts to distant servers, CatharinAI operates locally on your iPhone, iPad, or Mac, leveraging Apple Silicon and the latest advancements in MLX and Foundation Models. Every response, insight, and idea remains securely on your device, encrypted, ephemeral, and entirely yours.
Behind the scenes, CatharinAI employs a flexible runtime that seamlessly transitions between Apple Foundation Models and MLX-optimized LLMs, all optimized for low latency and battery efficiency. Regardless of your activities, whether writing, learning, or coding, CatharinAI adapts to your workflow without compromising your privacy. It’s intelligence that doesn’t demand your data in return.
Crafted with the same meticulous attention to detail as every vaic.at app, Catharina’s interface exudes a natural and dynamic feel across iOS, iPadOS, and visionOS. Its elegant chat view, model picker, and privacy modes are designed to make AI feel human—not intrusive. Catharina represents a significant step towards digital autonomy, offering a private, local, and deeply personal form of intelligence.
Coming soon, we will share more in early 2026.
vAI LIVE – PCC
When the user opts in, both CathAI and CatharinAI can access PCC (Private Compute Cubes) by CUEB.ES.
These Private Cubes are instances of an imitation of a generic user that collects data online, such as job listings, housing data, and salary comparisons, from publicly available or licensed data sources accessible through our systems.
After data collection is complete, the simulated user disappears, and the proxy shuts down.
Platforms that provide this information often deploy hidden tracking capabilities. Compute Cubes by CUEB.ES will ensure that data collection is a one-way street.
How PCC works
Private Compute Cubes (PCCs) are trained locally using large language models (LLMs) to be used as a proxy and simulate generic browsing behavior through randomized, anonymized profiles to prevent traceability or behavioral linkage, while modifying generic profiles to make them untraceable.
Our computers run these LLMs locally and have access. The requested data will be retrieved from the sources associated with our accounts without ever sharing the user’s personal information.
The proxy then transmits the data to CathAI and CatharinAI for inference purposes. Additionally, users have the option to view the raw data and its sources.
We are currently working hard on building PCC, more information about PCC will be released in 2026.
All PCC operations adhere to EU data protection and platform compliance standards, ensuring that data retrieval respects source terms and user consent.
During the alpha testing phase, PCC leverages locally deployed MistralAI large language models, enhanced with vaic.at’s proprietary NLP modules to simulate generic user interactions while maintaining complete data privacy.
As far as we know, App Intents, Apple Intelligence, and Apple Foundation Models will be capable of interfacing with PCC in future releases.
We intend to make this integration as seamless as possible, enabling our apps and AI features to extend naturally to Siri and the broader Apple Intelligence ecosystem once officially released by Apple.