Zane Chee is a computer-use engineer working with Cua and the creator of pi-computer-use. His software connects agents to desktop interfaces through inspectable observations and actions rather than requiring every application to expose an API.[3][5]
Start with pi-computer-use →
Education
B.Sc. in Computer Science, Software Systems (in progress; expected April 2027)
Aug 2023 – end date unverified
[4]Links
Computer Use Engineer (contract) · Cua
Codex Ambassador (unpaid volunteer) · OpenAI
Career
Chee’s experience includes e-commerce software at Hypotenuse AI, agent tooling at SEA Group and cofounding Vinesweeper. His Cua contract builds on maintaining pi-computer-use. He also serves as an unpaid Codex Ambassador, organizing events and demonstrations in Singapore. His résumé records computer-science study at SMU and a non-degree summer program at Oxford.[3][4]
Contributions
Observe before acting on a desktop
pi-computer-use provides tools for finding windows, reading interface structure, inspecting controls and taking actions. Immutable observations tie a proposed action to the state it was based on; progressive inspection lets an agent request more detail where needed. The Pi extension uses platform accessibility facilities and documents different capabilities across macOS, Windows and Linux. macOS setup requires Accessibility and Screen Recording permission; Linux behavior depends on the graphical session and accessibility support.[5]
Separate grounding from execution
Cua’s Jev announcement describes a Driver that observes, acts and verifies progress, with optional perception when structural interface information is insufficient. Candidate targets constrain what the executor can select. The demonstration reports timing and cost for a single 2048 run. That example illustrates the architecture’s intended efficiency under a bounded task.[6]
Tools, writing & talks
Tools
Start here
pi-computer-use
github.comA Pi extension exposing desktop observation, inspection and actions through platform-specific helpers.
Read the platform capability matrix and tutorial first. You need Pi and an interactive supported desktop; macOS also requires the documented system permissions.
Creator and maintainer[5]
Writing, research & talks
A team presentation of bounded target selection, execution and optional visual perception.
Cua computer-use engineer; team project[3][6]Original source