Sigil Wen, a self-taught coder who once lived in an AI hacker house with prominent researchers like Andrej Karpathy, has launched Underdog, an AI assistant positioned as the most private competitor to existing models such as Instinct and Muse. Backed by Silicon Valley investors including Patrick Collison and Andreessen Horowitz, Underdog operates entirely on-device, running on Macs and Windows PCs with plans for Linux, iPhone, and Android support. This on-device design ensures that user data stays local, enhancing privacy by avoiding centralized data processing. The assistant is powered by Husky, Wen’s own inference engine optimized to minimize data transfer within the computer hardware and securely encrypt user account keys.

Underdog utilizes a 27-billion parameter reasoning model fine-tuned from Qwen3.8-27B, which Wen claims matches or exceeds the performance of tools like Claude Opus 4.6 on certain benchmarks. While these smaller models may lack the scale of data center-hosted counterparts, Wen maintains they are sufficiently capable for everyday tasks like shopping research or math problem-solving without compromising privacy. He envisions the efficiency and capabilities of on-device models improving continuously, negating the need for users to trade personal data for AI performance. Wen’s AI manifesto underlines this ethos, questioning why AI use should require surrendering private information.

The business model for Underdog is innovative in eschewing traditional subscription or ad-based revenue. Instead, the startup takes a small fee from transactions made through the assistant using Stripe’s payment infrastructure. This model, inspired by fintech’s interchange fees, aligns the company’s incentives with user privacy, as Underdog does not need to monetize data but benefits when users conduct secure payments seamlessly. This approach contrasts sharply with many other AI assistants that rely on data collection to fuel advertising or third-party sales, a practice raising concerns given the sensitive nature of personal assistant data that often includes health, financial, and familial information.

Conway Research, the startup behind Underdog, has attracted a star-studded list of investors beyond Collison and a16z, including Khosla Ventures, SV Angel, and the Anthology Fund. Wen, a Thiel Fellow and veteran of pioneering AI work alongside notable figures in Silicon Valley, created Underdog initially for personal use to build a tool he would trust for his own family. The current invite-only beta release marks Underdog as a promising entry into the AI assistant market for users prioritizing strong privacy protections without sacrificing everyday usability.