Conway Research Launched Underdog AI Assistant

The startup introduced a new private, on-device AI tool currently available in an invite-only beta.

Updated on Oct. 6, 2026 in Artificial Intelligence

Bold vector editorial illustration showing a small metallic geometric lattice on a desk, representing private on-device artificial intelligence technology. — Conway Research has launched Underdog, a new private, on-device AI assistant designed to prioritize user data security by operating locally on personal computers. AI Illustration. Upload story photo >

Live Poll

Would you choose an AI assistant that promises full data privacy over one with more power?

Conway Research released an invite-only beta version of its new AI assistant, Underdog. The tool is designed to run entirely on-device to prioritize user data privacy.

Why it matters

Underdog aims to address growing concerns regarding data security by operating locally rather than on external servers. By keeping information on the user's machine, the startup seeks to provide advanced assistance without requiring the trade-off of personal data.

The Underdog model operates with 27 billion parameters and utilizes the proprietary Husky inference engine. This engine optimizes performance by minimizing data movement between the main and graphics processors.

The players

Sigil Wen

The founder of Conway Research who moved to Silicon Valley at age 17.

Conway Research

The Silicon Valley-based startup responsible for developing the Underdog AI assistant.

The details

Underdog functions locally on Windows PCs and Macs, encrypting access keys for connected accounts like email to ensure security. While the service is currently free and ad-free, Conway Research plans to monetize by charging a percentage on payment transactions facilitated through Stripe.

Sigil Wen launched the invite-only beta of Underdog on October 5, 2026.

The Tech Race

This development reflects a broader industry shift toward local, edge-based AI processing as an alternative to resource-heavy cloud infrastructure. By building on the Qwen3.8 27B model, Conway Research is positioning itself against cloud-native competitors by emphasizing local privacy controls.

Users can process sensitive account data locally without fear of external data scraping or server-side exposure. As the service expands, the integration of Stripe payment rails will allow users to conduct transactions directly through the assistant.

The takeaway

Underdog represents a pivot toward keeping personal data on a user's own machine rather than sending it to third-party clouds. Individuals interested in privacy-focused tools should monitor the platform's expansion to mobile and Linux operating systems.