Conway Research Secured Funding for On-Device AI

The startup aims to offload AI compute to local Apple Silicon with its newly announced Underdog assistant.

Updated on Oct. 2, 2026 in Artificial Intelligence

Isometric editorial illustration of a complex microchip die, representing the local compute architecture of on-device AI assistants. — Conway Research has raised capital from Andreessen Horowitz and other venture firms to expand its local AI assistant technology on Apple Silicon hardware. AI Illustration. Upload story photo >

Live Poll

Do you trust AI assistants more when they run entirely on your own personal devices?

Conway Research has secured investment from Andreessen Horowitz, Khosla Ventures, and Hummingbird VC to scale its local AI initiatives. The company launched Underdog, an AI assistant designed to operate locally on Mac and iPhone hardware.

Why it matters

By moving computation from cloud servers to local consumer hardware, Conway Research seeks to eliminate privacy risks while shifting operational costs to the end user. This approach leverages the specialized inference engines the company developed for the Apple Silicon ecosystem.

The company’s inference engines reportedly achieve 730 tokens per second on MacBook devices, a 4.5 times speed increase compared to Apple MLX. These models are optimized to reside entirely within device storage.

The players

Conway Research

A startup founded by Sigil Wen in 2026 that specializes in local AI inference engines for the Apple ecosystem.

Sigil Wen

The founder of Conway Research who previously gained notice for a Web 4.0 essay that received 10 million views.

Andreessen Horowitz

A venture capital firm that manages extensive portfolios in cloud infrastructure, consumer platforms, and artificial intelligence.

The details

Underdog operates by running models locally, ensuring data remains on the user's device rather than being transmitted to an external server. The system utilizes custom-built inference engines—software that executes trained AI models—designed specifically to maximize the performance of Apple Silicon processors. The 4-billion-parameter Woof model is optimized to maintain a small memory footprint while providing assistant capabilities.

2025: Sigil Wen served as a Thiel Fellow.
February 2026: Sigil Wen published a Web 4.0 essay.
2026: Conway Research was founded.
October 2, 2026: Sigil Wen announced the funding round.

The Tech Race

Conway Research is positioning itself against cloud-reliant LLM providers by proving that local hardware can handle complex inference tasks. The company is actively challenging the performance baseline established by the Apple MLX framework.

Users with Apple Silicon-equipped Macs and iPhones may gain the ability to run high-speed AI assistants without ongoing cloud subscriptions or privacy exposure. Deployment timelines and availability for specific devices have not yet been provided by the company.

The takeaway

Conway Research represents a shift toward edge-based AI where the user's hardware serves as the primary compute engine. Interested parties should monitor the company for future benchmarks comparing its custom inference speed against standard open-source local LLM implementations.