A desktop AI agent approach aimed at cutting token cost
This desktop AI agent avoids the approach, where an AI controls the mouse and keyboard directly. It instead uses an operating-system-based way to connect with s.
The goal is not to support a handful of apps, but to scale to thousands of s. The team sees as too slow and too expensive, especially if each user could require around $300 in token costs.
The product focuses on processing and aims to run smoothly at high TPS. The idea moved from to MVP and then to after about five months of work on .
Key points
- The approach avoids because it is described as slow and costly.
- The team wants to avoid a model where each user may consume about $300 in token costs.
- The target is with thousands of s, not just a small list.
- The product has moved through , MVP, and .
- The focus is processing and smoother operation at high TPS.