Conway Research Released Underdog AI for Apple Silicon

The new AI assistant runs locally on Mac and iPhone hardware to prioritize user privacy and speed.

Updated on Oct. 2, 2026 in Artificial Intelligence

Isometric editorial illustration of a silicon semiconductor wafer with intricate geometric golden circuit pathways on a dark matte surface. — Conway Research has launched Underdog AI, a new assistant that operates entirely offline on Apple silicon devices to improve user privacy. AI Illustration. Upload story photo >

Live Poll

Do you prefer using AI tools that process data locally on your device instead of the cloud?

Conway Research has launched its Underdog AI assistant, designed specifically for Apple silicon devices. The software operates entirely offline, ensuring that all user data remains on the local device.

Why it matters

By processing tasks locally without a Wi-Fi connection, Underdog offers a privacy-focused alternative to cloud-based AI assistants. This local-first approach mitigates data security concerns associated with sending sensitive information to external servers.

The system utilizes a 4-billion-parameter model named Woof powered by the Husky inference engine. In testing across 16 unique tasks, the model achieved speeds of 730 tokens per second on an Apple M5 Max chip.

The players

Conway Research

This technology firm specializes in the development of artificial intelligence models and inference engines.

The details

The Underdog assistant uses the Husky inference engine to execute complex tasks directly on Apple hardware. Because the software functions without internet connectivity, it provides instantaneous processing while keeping all personal data contained within the device hardware.

October 2, 2026: Conway Research published the benchmark results for its new AI software.

The Tech Race

Comparing the performance of Underdog AI against the Apple MLX framework highlights the ongoing optimization race for local artificial intelligence models. This advancement signals a broader industry shift toward running sophisticated models directly on consumer hardware rather than relying on external cloud processing.

Users can now leverage powerful AI capabilities without worrying about data privacy or needing an active internet connection. This provides a more responsive workflow for Mac and iPhone owners, though performance speeds will vary depending on the specific Apple silicon chip installed in their device.

The takeaway

The ability to run a 4-billion-parameter model locally marks a significant improvement in on-device AI efficiency. Consumers interested in privacy should prioritize software that processes tasks locally, as it prevents the exposure of personal information to third-party cloud servers.