Urgent.News

What's breaking now, across thousands of outlets.

AI

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

Sigil Wen, backed by a Silicon Valley's who-who, has built an on-device AI assistant that promises to be free, fully private, and capable for everyday tasks.

Self-taught coder Sigil Wen, once a resident of an AI hacker house in Silicon Valley alongside AI luminary Andrej Karpathy, has launched Underdog, a new private AI assistant. Wen's journey includes collaborations with notable figures like Perplexity founder Aravind Srinivas and OpenAI researcher Noam Brown, testing early AI tools such as a chatbot that later evolved into Claude and an image generator that became Midjourney.

Wen's fascination with AI engineering led him to optimize GPT-2 to run on an Apple Watch. Wen's latest venture, Underdog, marks a departure from conventional AI assistants' data practices, operating entirely on users' devices, including Macs and Windows PCs. This approach ensures the user's data remains private, as it does not leave their personal devices.

Underdog incorporates encryption to secure access to the user's email and other accounts. Wen's model, running a 27-billion parameter reasoning model fine-tuned from Qwen3.8 27B, is competitive with top-notch AI models available today. Underdog aims to deliver comparable performance to Claude Opus 4.6 in various tasks, such as shopping research and math homework assistance.

The startup, Conway Research, boasts an impressive list of investors, including Naval Ravikant's Angel investor, Patrick Collison, and prominent names like Andreessen Horowitz, Khosla Ventures, and Hummingbird.

Written by urgent.news from TechCrunch's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at techcrunch.com →

More in AI

Getting to Reliable AI-Driven Development

To truly transform software delivery with AI, organizations must embrace a spec-driven approach. By grounding AI in clear specifications, this approach prevents AI from generating hallucinations while optimizing token costs by 8X–12%. The use of AI in software development is now practically universal.

More from Tuesday 6 October →