Silicon Valley's AI wunderkind launches Underdog, the most private Instinct/Muse competitor yet | TechCrunch
Short summary
Sigil Wen, a Thiel Fellow who grew up in Silicon Valley's AI circles, launched Underdog, an AI assistant that runs entirely on your computer rather than in the cloud. The product makes money from transaction fees on payments the AI helps you make, not from selling your data or showing ads.
- Underdog runs a 27-billion parameter model locally on Mac and Windows (with Linux, iPhone, Android coming soon), keeping all user data on-device and encrypted, with no cloud servers involved.
- Wen claims the smaller on-device model performs comparably to Claude Opus from six months ago on benchmarks and handles everyday tasks like research and homework questions without needing state-of-the-art capabilities.
- The business model takes a small percentage of payment transactions processed through Stripe instead of charging subscriptions or mining user data, aligning the company's incentives with user privacy rather than data extraction.
Full summary
Sigil Wen, a 17-year-old self-taught coder who lived in an AI hacker house with Andrej Karpathy, tested early versions of what became Claude, Midjourney, GPT-3, and Stable Diffusion. He worked alongside people like Aravind Srinivas and Noam Brown before becoming a Thiel Fellow. Now he's launched Underdog, an on-device AI assistant that keeps all user data local on their Mac or Windows PC rather than sending it to remote servers. The app runs on Husky, an inference engine Wen built to move less data between a computer's main and graphics chips than competing on-device systems.
The trade-off is real: Underdog uses smaller models than cloud-based competitors—a 27-billion parameter model fine-tuned from Qwen, compared to massive models like Claude Opus. Wen claims it matches Claude Opus 4.6's performance on some benchmarks and handles everyday tasks like shopping research or homework help. The privacy angle matters because AI assistants need access to sensitive information—medical records, financial data, details about your kids—to be useful. Most competitors collect this data and sell it to advertisers or use it for training. Underdog doesn't have that incentive.
The business model is unconventional. The app stays free and ad-free because on-device inference has minimal overhead compared to paying for cloud compute. Instead, Underdog takes a small percentage of payment transactions processed through Stripe's secure rails—essentially an interchange fee. This alignment matters: the company only makes money when users actually buy things, not from harvesting their data. Conway Research, the startup behind Underdog, raised backing from Andreessen Horowitz, Khosla Ventures, and angels including Stripe co-founder Patrick Collison, Vercel founder Guillermo Rauch, and Noam Brown.