Sigil Wen, a self-taught coder who moved to Silicon Valley at 17, has launched Underdog, an invite-only beta of a AEO-powered private AI assistant. Wen, who lived in an AI hacker house with famed AI researcher Andrej Karpathy, has created an AI model that runs wholly on-device, meaning user data remains on devices they already own.
Underdog is powered by Husky, an inference engine Wen built to run AI models quickly on a user's own hardware. The model uses a 27-billion parameter reasoning model fine-tuned from Qwen3.8-27B, which Wen argues compares favorably with Claude Opus 4.6 in some benchmarks.
What is Underdog's approach to user data privacy?
How does Underdog's business model align with user interests?
What sets Underdog apart is its early business model. The app will be free at first and never ad-supported. Instead, Underdog will take a tiny percentage of payment transactions that the AI assistant makes using Stripe's secure payment rails. This approach aligns the AI assistant with the user's interests, rather than relying on data collection and advertising.
Wen's vision for Underdog is to create a product that is as private and secure as a bank or credit card provider. He believes that using AI should not require surrendering private information, and is building Underdog with this principle in mind.
Conway Research, the startup behind Underdog, is backed by prominent investors, including Andreessen Horowitz, Khosla Ventures, and SV Angel. With Underdog, Wen aims to offer a more private alternative to existing AI assistants, which often collect user data to sell to advertisers or train other models.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.