From an idea to a reviewed change.
See the workflow.
Give your AI a task, keep the party in the loop, and review the combined work before anything is kept.
- Step 01 · Set up
Pick a project. Choose an AI for each seat.
Start a brand-new project, or pick a Git project with no unsaved changes. CodeParty won’t start on a project with unsaved work, so nothing of yours is stashed or thrown away.
Each seat uses Codex, Claude, Gemini or GitHub Copilot through the command-line tool that person already signed in to. A short checklist shows anything missing and the one step that fixes it.
- Step 02 · Split
Everyone gets their own copy.
CodeParty gives each person a fresh branch and a separate copy of the project, so nobody’s edits touch the shared project or anyone else’s copy.
Teammates can watch each other’s files read-only. The editor works offline.
- Step 03 · Coordinate
Everyone knows what everyone else is doing.
File claims, AI messages and shared agreements show up in one activity stream, so the party sees a likely collision before it happens. A claim is a heads-up, not a lock.
Before each turn, every AI gets a short briefing: what the party decided in Party memory, what teammates are working on, and files you both changed. If one AI removes something a teammate’s code still uses, that teammate gets a heads-up.
- Step 04 · Review
Try everyone’s work together first.
Review changes combines everyone’s committed work in a temporary copy and, after asking you, runs your project’s checks there. Nothing touches the shared project.
If a check fails, Review names the check, the files it mentions and whose changes touched them. Check commands run in a copy of the project, but they are not sandboxed.
- Step 05 · Keep
Keep it only when it passes.
Keeping work needs a passing trial merge, no known conflicts, a clean shared project and a person’s confirmation. If any of those fail, nothing is kept.
Gitignored files such as .env in the host’s project are left alone, and work in which one seat changed another seat’s copy is refused.
What changes when your AI joins a party.
Single-player AI tools are good at helping one person. CodeParty adds what a group needs on top of the AI each friend already uses.
- Which AI
- On your own
One AI helps one person. Friends who use different AIs each work alone.
- In a party
Each friend brings Codex, Claude Code, Gemini or GitHub Copilot, signed in on their own Mac, to the same party.
- Where edits go
- On your own
Into your folder. Sharing usually means pushing and pulling by hand.
- In a party
Into that friend’s own copy. The shared project changes only when the party keeps work.
- Knowing what friends are doing
- On your own
Your AI only knows what you tell it.
- In a party
Before each turn, every AI gets a short briefing on teammates’ work and party decisions, plus a heads-up when a teammate removes or renames something its code uses.
- AIs working together
- On your own
Each AI works alone.
- In a party
AIs can send each other typed questions, and teammates can hand off unfinished work with a note. The party sees both.
- Combining the work
- On your own
Merge it yourself and hope the checks still pass.
- In a party
Review changes tries everyone’s work together in a temporary copy and runs the project’s checks first.
- Deciding what to keep
- On your own
Whoever pushes decides.
- In a party
Keeping work, or undoing the last keep, needs a yes from more than half of the party. Anyone can propose.
- Where the code lives
- On your own
Depends on the tool: your computer or the tool’s cloud.
- In a party
On the Macs in the party. The host’s Mac runs it; there is no CodeParty cloud workspace.
Good to know: CodeParty runs on Macs with Apple silicon and macOS 14 or later, holds up to four people, and needs the host’s Mac to stay on. Friends watch each other’s files rather than type in the same file together.