Rabbit Founder Lu Cheng: People Should Not Change How They Think for Agents

Summary

Rabbit founder Lu Cheng says the company’s new OS3 is intended to make personal AI agents easier to use by letting people state an objective in ordinary language instead of sorting tasks into separate work and personal categories. OS3 can be instructed from the web or Telegram, use a user-selected AI model, and connect up to five existing computers to access their files, applications, and work environments. The system keeps a single continuing conversation and is expected to decide which device and tools should handle each request. Rabbit sees local execution as a privacy boundary: files can remain on the selected computer rather than being uploaded, although the cloud model still participates and users must install software and grant permissions. The company uses a bring-your-own-key model, allowing users to supply API keys from different model providers rather than locking Rabbit to one vendor. Lu Cheng says model competition has no settled winner, so improvements from any provider could benefit Rabbit’s product. He also acknowledges unresolved barriers, including potentially high token consumption, the technical difficulty of logging into websites from unfamiliar remote environments, and the gap between a simple interface and the complexity required behind it. OS3 follows criticism of Rabbit’s R1 device, which received a 3/10 early review from Wired; Lu says more than 100,000 R1 units have been delivered globally and that the return rate is below 5%. Rabbit is also developing Cyberdeck, a portable computer intended to serve as a unified task entry point while calling on users’ home or office machines. Lu argues that the company is pursuing an Agent operating environment rather than merely shifting from hardware to software, but he also says the long-term question is whether ordinary people will find personal agents genuinely more convenient than performing simple tasks themselves.