# Former CoScreen team builds open-source pair programming app Petal in under eight weeks

> Petal is a free, open-source screen sharing app for pair programming in which shared windows appear as movable windows on the other person's desktop, and collaborators can click in to type and control them.

- Project: Petal by thedracle
- Kind: app
- Built with: Fable, Astra
- Noticed: 228 points on Hacker News (when we found it)
- Added: 2026-11-21T10:07:58+00:00
- Web page: https://ai-news.ikangai.com/built/petal/

Petal runs on macOS and Windows, and a web app is also available. Shared windows appear as real, resizable windows on your desktop instead of tiles in a grid. Several people can share at once, each with a labeled cursor. Anyone can click into a shared window to type, scroll, and click. Rooms persist, so a team can join a room such as eng-sync with one click instead of passing links around. Media runs over LiveKit. Users can join the hosted instance with no setup, or point the app at their own LiveKit server using a docker-compose file. Linux support is promised next.

The project comes from four people who worked on CoScreen, a screen sharing tool that Datadog shut down on July 31. The author, Jason Thomas, writes that the first version of Petal took under eight weeks to build from scratch, with frontier models like Fable and Astra doing much of the work. He says real-time video and window capture was some of the most unforgiving code the team has written, and they did not expect that. The post does not say how the work was split between the people and the models. The authors say a fuller writeup on what worked and what didn't is coming.

## In the author's words

> The first version of Petal took under eight weeks, from scratch, with frontier models like Fable and Astra doing a lot of the work.

## What to take from it

Keeping the hard media layer out of the core product is the lesson here: Petal uses LiveKit for video and lets users swap in their own server, so the team only had to build the app itself. Expect real-time capture code to be the slow part even when the rest moves fast.

## See it

- [Petal](https://petal.live/)
- [Discussion on Hacker News](https://news.ycombinator.com/item?id=50092470)

---

AI wrote this entry from the project's own README or its author's post. That AI built the project is the author's statement, quoted here. Nobody at IKANGAI ran or tested it.
