Analysis
Sigil Wen, a self-taught coder and Thiel Fellow, launched Underdog on Monday, TechCrunch reported, a privacy-focused AI assistant that runs entirely on a user's own device rather than in the cloud. Wen's path to the launch includes living in an AI hacker house with researcher Andrej Karpathy at age 17 and working alongside future founders Aravind Srinivas, now of Perplexity, and Noam Brown, now at OpenAI. Underdog's parent company, Conway Research, is backed by Andreessen Horowitz, Khosla Ventures, Hummingbird, SV Angel and Anthology Fund, with angels including Stripe co-founder Patrick Collison.
The product runs on Husky, an inference engine Wen built to run a 27-billion-parameter reasoning model — based on Qwen3.8-27B — efficiently on consumer Macs and Windows PCs, with Linux, iPhone and Android support planned. Wen claims the model compares favorably with Claude Opus 4.6 on some benchmarks, a notable claim for a model roughly two orders of magnitude smaller than frontier closed models.
Privacy as the wedge against Instinct and Muse
Underdog's pitch is explicitly against Instinct, the invite-only AI assistant that jumped from a roughly $100 million to a $10 billion valuation in weeks this year, and Muse — both of which monetize partly by collecting user data.
Underdog launches free and ad-free, earning revenue instead through small transaction fees on Stripe-processed payments the assistant initiates — an interchange-style model that only works if the assistant is actually transacting on users' behalf at meaningful volume, not just answering questions.
Running inference locally keeps Underdog's costs low since there's no cloud GPU bill per query, which is also what makes the ad-free, no-subscription model viable in the first place. But local inference caps capability at whatever a consumer's own Mac or PC can run — Instinct and Muse can scale compute per request in the cloud in ways Underdog structurally cannot, which matters most for complex, multi-step agent tasks rather than simple Q&A.
Whether 'runs on your machine, keeps your data' is a strong enough wedge against venture-backed incumbents with hundreds of millions in funding and a head start on agent capability is the open question. Underdog's bet is that enough users will trade some capability ceiling for genuine on-device privacy — a trade Apple has made central to its own AI pitch, suggesting Underdog may ultimately be competing with Apple's roadmap as much as with Instinct or Muse.