Privacy accounts without monthly subscriptions: how architect “Husky” flips the smart-assistant business model

Technical norm settled over the past two years on a quasi-inevitable equation that forces the user into one of two trade-offs: either surrendering data privacy to data centers that train models and sell their behavior to advertisers, or paying a monthly subscription that covers the high inference cost of cloud servers. Breaking this constraint was the premise on which Seegil Win, a colleague of the Thiel program who began his software career early in California and worked alongside leading AI researchers and tested early versions of tools like Claude and Midjourney, announced the launch of the beta version of his smart assistant “Under Dog”.

The new assistant, developed by Conwy Research, is built on a purely edge-computing architecture that keeps all data entirely on the user's personal hardware on macOS and Windows, with plans to later expand to Linux, iPhone and Android. Win relies on an inference engine he created called “Husky” to reduce data movement between the CPU and GPU compared with competing edge engines, encrypting account keys and email that the user authorizes the assistant to access. The system runs an inference model comprising 27 billion derived parameters, tuned through specialized training from “Qwen 3.8-27B”, and its founder claims it matches the performance of “Claude Opus 4.6” on some benchmark metrics that represented top performance six months ago, making it sufficient for daily tasks such as market research and solving mathematical problems.

The deeper bet in this move lies in cost engineering, not in technical specifications alone.Freeing the assistant from cloud inference servers removed the massive operational financial burden that weighs on model developers and forces them to impose recurring subscriptions. Win borrowed a model from the fintech sector, backed by an angel investment from Patrick Collison, co-founder of Stripe, along with funding from Andreessen Horowitz via partner Chris Dixon, and from Hustle Ventures, the “Anthology” fund jointly managed by Menlo Ventures and Anthropic, and angel investors including Guilherme Rausch and Naom Brown. The business model provides the service for free without ads, taking a small percentage of the financial transactions and purchases that the assistant executes through Stripe’s secure payment channels, tying the company’s revenue to facilitating commerce rather than profiting from the exploitation of individuals’ data.

The editorial board sees this shift opening a strategic window for work teams and companies in the Gulf, Egypt and the Levant.In regional business environments, especially banks and health institutions subject to stringent regulations that limit data export to foreign clouds, the edge model eliminates the compliance bottleneck of digital sovereignty without the need to purchase costly server farms for everyday use cases. For entrepreneurs and developers, shifting the computing cost to the client’s hardware completely changes the unit economics of emerging products, moving budgets from covering onerous API fees to investing in locally improving model efficiency, and the skill in demand in the regional job market becomes edge computing and transaction compression rather than merely calling ready-made APIs.

The launch of “Under Dog” reveals that the gap between central computing capabilities and personal-device computing is shrinking at a pace that exceeds expectations, putting companies that built their business models on selling cloud subscriptions under genuine pricing pressure. When a desktop computer can perform inference tasks efficiently at an operating cost that approaches zero for the developer, the logic of AI monopoly behind centralized servers begins to crack, leaving the decision to users to regain control over their privacy and data.