Underdog is entering the AI assistant market with a different economic and privacy model from many cloud-based competitors.
The supplied source says the assistant runs entirely on the user’s device, currently on Macs and Windows PCs, with Linux, iPhone and Android versions planned.
Underdog is powered by an inference engine called Husky and currently uses a 27-billion-parameter reasoning model fine-tuned from Qwen3.8-27B.
Running locally changes the business model. Because the company does not need to pay ongoing cloud inference costs for each user interaction, founder Sigil Wen says the product can initially be free and remain ad-free.
Instead, the company plans to take a small percentage of payment transactions completed by the assistant through Stripe’s payment rails.
That creates an interesting alignment: the company makes money when the assistant successfully completes commerce rather than by monetizing user attention or personal data.
The trade-off is capability. Smaller local models may be cheaper and more private, but they may not match the breadth of frontier models hosted in large data centers.
Conway Research is backed by investors including Andreessen Horowitz, Khosla Ventures, Hummingbird, SV Angel and the Anthology Fund.
BTI’s bottom line: Underdog is testing whether privacy and lower inference costs can become competitive advantages in consumer AI. If on-device models become capable enough for common tasks, the assistant market may support very different economics from today’s subscription-heavy cloud model.