Inspiration

I was inspired by Dominik Kundel's talks (at AI Engineer World's Fair) about building on the Codex app server! As soon as I got to the airport I to fly home from the conference, I began to build multAIplayer to enable trusted teams to build with Codex, together.

What it does

multAIplayer is built on the Codex app server, with a hosted (or self-hosted, both options are available) relay connecting multiple users to enable them to work on the same project together. One person hosts the room using their Codex process running on their local machine. Through the relay, others can join and work on the same project. multAIplayer has both a macOS app and a CLI.

The app contains a Monaco editor, xterm,js terminals, diffs, GitHub Actions, an in-app browser, the option to use Cloudflare Quick Tunnels to share local development builds with others in the room, and of course, chatting and building with Codex and your trusted team.

The CLI is similar, but doesn't have the extra features that the app has.

The app and CLI are divided into "teams" and each team can build in "rooms" (basically separate projects connected to GitHub projects or local folders on the host's machine.

If the host of the room ever runs out of usage, they can hand off their role using a "Host Handoff" package, and transfer authority to someone else who has Codex running on their local machine.

Before anything is run on a host's machine, or through their Codex, they have to approve Codex turns. Anyone can suggest a Codex turn, but the host must review the acceptance packet and either accept or decline the action.

How we built it

multAIplayer was built entirely using GPT-5.6 Sol Medium, in Codex itself. Basically, it was built with Codex, for Codex, in Codex, on the Codex app server, for people to use it to build with Codex, together.

Challenges we ran into

I wanted to make a very security-sensitive product. I'll admit, the first iteration of multAIplayer has some hand-rolled crypto involved on the security side. I quickly realized that was not ideal, and instead incorporated RFC 9420 MLS via mls-rs, and P-256 HPKE/signatures for the cryptographic side of things.

Accomplishments that we're proud of

I am phenomenally excited to have built a project that I'm proud of, and that I feel fills a niche.

What we learned

In building multAIplayer, I learned a lot about code quality, repository management, and just plain building a project that people will hopefully use.

What's next for multAIplayer

multAIplayer is free and open source, and I'd love to see it become a project people depend on!

Built With

  • codex
Share this project:

Updates