Silicon Valley’s AI wunderkind launches Underdog, the most private Instinct/Muse competitor yet

Sigil Wen, a Thiel Fellow and self-taught coder from Silicon Valley’s early AI scene, launched an invite-only beta of Underdog, an on-device AI assistant that runs wholly on users’ machines and emphasizes privacy. The app uses Wen’s Husky inference engine and a 27-billion-parameter model fine-tuned from Qwen3.8-27B, and the company says it will be free and ad-free, funded via tiny payment fees on transactions routed through Stripe.

By AI Newsroom· Reviewed by Pranav, Founder & Editor-in-ChiefPublished 1 minute agoUpdated 1 minute ago0 views

Silicon Valley’s AI wunderkind launches Underdog, the most private Instinct/Muse competitor yet

Why It Matters

Underdog represents a push toward locally run AI assistants that keep user data on-device rather than sending it to cloud providers, which could change privacy trade-offs for everyday AI tasks. Its investor backing and novel payments-based revenue model suggest a commercially viable alternative to data monetization used by some AI services.

Key Facts

Founder: Sigil Wen
Company: Conway Research (product: Underdog)
Launch: Invite-only beta launched Monday (source article date)
Fellowship: Thiel Fellow
Inference engine: Husky (built by Wen)

Sigil Wen, who moved to Silicon Valley as a teenager and worked alongside early members of the current AI landscape, has opened an invite-only beta of Underdog, an AI assistant designed to run entirely on users’ devices. Wen, now a Thiel Fellow, says Underdog presently supports macOS and Windows, with Linux, iPhone, and Android versions planned. The app encrypts keys to users’ connected accounts and keeps data on the hardware the user already owns. Underdog’s local performance relies on Husky, an inference engine Wen developed to reduce data movement between a computer’s CPU and GPU. The assistant runs a 27-billion-parameter reasoning model that Wen describes as a fine-tune of Qwen3.8-27B. He contends that, on some benchmarks, the model holds up against Claude Opus 4.6 or what was considered top-tier performance six months ago, and that it can handle common assistant tasks like shopping research and homework help. Wen emphasizes privacy as a central design goal. Because the models execute on-device, user data does not need to be streamed to remote servers for inference. Underdog also encrypts credentials for services users authorize it to access. Wen frames the approach as avoiding the trade-off between capability and privacy, writing in an "AI manifesto" that users should not have to surrender private information to use AI. Conway Research, the startup behind Underdog, is backed by a roster of well-known investors. Angel backers include Stripe co-founder Patrick Collison, Vercel founder Guillermo Rauch, and researchers such as Noam Brown. Institutional investors named in the article include Andreessen Horowitz (via partner Chris Dixon), Khosla Ventures, Hummingbird, SV Angel, and the Anthology Fund. Wen says the company will offer Underdog for free and ad-free, and plans to monetize by taking a small percentage of payment transactions the assistant facilitates using Stripe’s payment rails, rather than charging subscriptions or selling user data.