Silicon Valley’s AI wunderkind launches Underdog, the most private Instinct/Muse competitor yet
Sigil Wen, a Thiel Fellow, has launched Underdog, an AI assistant designed for on-device operation to ensure user privacy. The app runs entirely on the user's hardware, processing data locally and encrypting sensitive account information.
Intelligence analysis by Gemini 2.5 Flash Lite
Underdog, an AI assistant from Thiel Fellow Sigil Wen, emphasizes user privacy by running entirely on-device, meaning data stays local. It uses smaller, efficient models and plans a unique revenue model based on transaction percentages via Stripe, avoiding data mining and subscriptions.
Imagine your phone or computer is like a super-smart notebook. Instead of sending your secrets to a big company's computer far away, this new AI app, Underdog, keeps all its thinking and your notes right inside your own notebook. It's like having a private diary that never leaves your room, and it makes money by helping you buy things, not by selling your secrets.
Analysis
Sigil Wen's Genesis
Sigil Wen's journey into the AI landscape began at a young age, marked by his immersion in Silicon Valley's vibrant AI hacker house culture. At just 17, he collaborated with future leaders in the field, including figures like Andrej Karpathy and OpenAI researcher Noam Brown. This formative period saw him experimenting with nascent AI technologies that would later become household names, such as Anthropic's Claude and OpenAI's GPT-3. His early work also caught the attention of prominent investors; he was hired by Naval Ravikant for Airchat, a social networking venture. Wen's technical prowess was further demonstrated by his ability to run complex AI models, like GPT-2, on unconventional hardware such as an Apple Watch. This background, characterized by hands-on experimentation and proximity to AI's pioneers, has clearly shaped his vision for Underdog.
Underdog's Privacy Architecture
The core differentiator for Underdog is its unwavering commitment to user privacy, achieved through a fully on-device operational model. Unlike many contemporary AI assistants that rely on cloud-based servers, Underdog processes all data locally on the user's Mac or Windows PC, with support for Linux, iPhone, and Android planned. This architecture inherently prevents sensitive user information from leaving their personal devices. Furthermore, the application incorporates robust security measures, including end-to-end encryption for access keys to authorized accounts like email. The engine powering this is Husky, Wen's custom-built inference engine, which is optimized for efficient AI model execution on local hardware, reportedly minimizing data transfer between a computer's main and graphics chips.
A Novel Business Model
Underdog's approach extends to its innovative business model, which eschews traditional subscription fees and ad-based revenue streams. Wen aims to generate revenue by taking a small percentage of transaction fees processed through the AI assistant, leveraging Stripe's secure payment infrastructure. This model, inspired by fintech practices, aligns Underdog's financial incentives directly with the user's transactional activity rather than their personal data. This is a stark contrast to competitors who often monetize user data through advertising or by using it to train future models. Wen's philosophy, articulated in his AI manifesto, questions the necessity of surrendering private information for AI utility, positioning Underdog as a tool built on trust and user alignment, akin to a bank or credit card provider.
Key points
- Underdog is a new AI assistant focused on user privacy through on-device processing.
- The app runs locally on user hardware, ensuring data remains private and encrypted.
- It utilizes a custom inference engine, Husky, for efficient on-device AI model execution.
- Underdog plans to monetize via a small percentage of transaction fees, avoiding subscriptions and ads.
- The startup, Conway Research, is backed by prominent investors like Andreessen Horowitz and Khosla Ventures.
Underdog could redefine user expectations for AI privacy, encouraging a shift away from data-hungry models towards more secure, on-device solutions. Its unique transaction-based revenue model might prove highly scalable and sustainable, offering a privacy-first alternative that attracts users wary of current data practices.
The reliance on smaller, on-device models might limit Underdog's capabilities compared to larger, cloud-based AI systems, potentially hindering its performance on complex tasks. Furthermore, the success of its novel transaction-fee business model is unproven and could face challenges in adoption or revenue generation.
