[Photo: Shutterstock]

Conway Research has launched an invite-only beta of Underdog, an AI assistant that runs only on a user's device. TechCrunch reported on Oct. 6 that Underdog currently runs fully on-device on Macs and Windows 11 PCs. Conway Research is also preparing versions for Linux, iPhone and Android.

Underdog is focused on privacy. It encrypts account keys for services such as email that users allow it to access, and processes AI inference on the device rather than in a data centre. Founder Sijil Wen (시질 웬) developed its own inference engine, Husky, for that purpose. Husky focuses on boosting speed by reducing data movement between a computer's CPU and GPU.

Underdog currently uses a 27 billion-parameter inference model that fine-tunes Qwen 3.8 27B. The model is smaller than the latest large models that run in data centres, but Wen said it showed performance comparable to Claude Opus 4.6 on some benchmarks. He said it can handle everyday tasks such as shopping research and math homework questions.

ㆍ[Tech Insight] "On-device AI has come to shake up an AI landscape centred on frontier models"

Conway Research is also taking a different approach to its revenue model. Underdog will be offered free initially and will not carry ads. Because the AI runs on users' devices, inference costs are not large. Instead, Conway Research is exploring a plan to take small fees on transactions that the AI assistant processes on the Stripe payments network. Stripe co-founder Patrick Collison is an angel investor in Conway Research.

Wen taught himself to code, moved to Silicon Valley at 17 and lived in an AI hacker house with Andrej Karpathy. There he worked with Aravind Srinivas, founder of Perplexity, and OpenAI researcher Noam Brown, among others, and he is now participating in Peter Thiel's Thiel Fellow program. Conway Research has received investment from Andreessen Horowitz, Khosla Ventures, Hummingbird, SV Angel and Anthology Fund, among others.

Keyword

#Conway Research #Underdog #Stripe #Patrick Collison #Husky
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.