a16z led Conway Research’s seed financing for local AI assistant Underdog

Article

a16z Betting on local AI assistantUnderdog: Email and calendars stay on your own computer

Beating AI news, Conway Research has launched personal AI assistant Underdog, backed by a16z, Khosla Ventures, and other institutions. It currently runs directly on Mac, connects to users’ email and calendars, and helps find emails, organize schedules, and handle everyday tasks. The model runs locally, so personal data no longer needs to be handed over to cloud AI processing, and it can continue to be used offline.a16z has confirmed that it led Conway Research’s seed round.

Underdog can run on an ordinary Mac, partly because the model has become small enough. The currently public Woof has 4B parameters, and its 4-bit weights are approximately 2.37GB. On the other hand, the team developed the Husky inference engine specifically optimized for Woof and Apple Silicon. Official tests were conducted on M5 Max; the same model was faster than 16 on all MLX tasks, reaching up to 4.5 times faster.

Founder Sigil Wen believes that everyday tasks such as finding emails, organizing files, and managing schedules do not require calling the most powerful model in a data center every time. As long as a small model is capable enough and fast enough, it can use the user’s own computer to complete many tasks directly.

Underdog also plans to expand to iPhone, Windows, Linux, Android, and NVIDIA devices. Wen calls this path “Underdog’s Law”: frontier AI that require data centers to run today may have an opportunity to enter personal devices in roughly months.

The team is also exploring scenarios in which Agent directly shops for users and more, with the goal of ultimately turning local AI into a personal Agent that retains personal context over the long term and handles tasks directly for users.

Original link https://m.theblockbeats.info/flash/370139

Key points

Conway Research has launched personal AI assistant Underdog, backed by a16z, Khosla Ventures, and other institutions.
Underdog currently runs directly on Mac, connects to users’ email and calendars, and helps find emails, organize schedules, and handle everyday tasks.
The model runs locally and can continue to be used offline.
The public Woof model has 4B parameters, and its 4-bit weights are approximately 2.37GB。
The team developed the Husky inference engine optimized for Woof and Apple Silicon.
Official tests on M5 Max reportedly found that the same model was faster than 16 on MLX tasks, reaching up to 4.5 times faster.
a16z has confirmed that it led Conway Research’s seed financing.
Underdog plans to expand to iPhone, Windows, Linux, Android, and NVIDIA devices.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

It handles email, calendars, and other tasks on the user’s own Mac, relying on a relatively small model and a specially optimized inference engine so that some AI assistant tasks do not need a cloud connection.

Why it matters to readers

This shows that small models and on-device inference can support personal Agent tasks such as email and scheduling, and may also drive the development of local AI applications and inference optimization.

For ordinary users, the key point is that data can remain on their own computer and the system can still be used offline; however, actual performance depends on model capability and product stability.

Risks and unknowns

The materials do not provide the amount of the initial funding round.
Funding information, performance data, and product capabilities still lack verification through a funding announcement, original investor statements, or independent sources.
The actual release dates for the iPhone, Windows, Linux, Android, and NVIDIA versions are unknown.
The materials do not specify the authorization methods or security mechanisms for email and calendar data.

a16z led Conway Research’s seed financing for local AI assistant Underdog

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center