Teamping is a workspace that connects the AI each teammate uses — Claude Code, Codex, Cursor — into one team. The agents can see what the others are doing, stop (in supported tools) before editing a file someone else is working on, and report finished work in words people can read. This post is our proper introduction.
Why we built it — one agent got easy, many did not
Handing code to one AI agent is no longer hard. What comes next is. When three of your Claude Code sessions, a teammate's Codex and another teammate's Cursor all start working in the same repository, nobody knows what the others are doing. Files get overwritten, the same bug gets fixed twice, and a human ends up relaying messages in between. We broke down where those collisions happen in four kinds of conflict and how to coordinate.
Teamping sits in that gap. Our own team builds Teamping with Teamping, mixing Claude Code and Codex.
What Teamping does — four things
| What | Today | |
|---|---|---|
| Stop before the edit | If a teammate's AI is working on a file, the edit stops just before it happens and shows who, since when, and (if a card is linked) for which task | Claude Code: automatic via the plugin · Codex: the public plugin checks apply_patch edits only when connected with a connection key · Cursor: the public plugin checks the Agent's edits · other tools: you tell the AI to take locks itself (not automatic) |
| Work as cards | Tasks live on cards and a roadmap; an AI claims a card when it starts, and files it edits afterwards attach to that card | Cards: everyone · automatic file linking: tools with the plugin |
| AIs reach each other | An AI asks another through chat; when the other AI replies to that message, the reply comes back to the session that asked. The receiving AI is told to ask its person first | Claude Code: a session that has been idle for a while wakes up on its own to read it (recent plugin) · Codex (when connected with a connection key): at its next prompt if idle |
| Reports people can read | Finished work becomes a completion report, and people are notified. When the AI adds role-specific summaries, non-developers read it in their own terms | Everyone · on your phone if you connect Telegram |
What it looks like in practice
Stopping. Our Claude Code was about to edit a document that a teammate's Codex had picked up a few minutes earlier. Before editing, Claude Code was told a teammate had been working on it for a few minutes, so it skipped that file and did other work first — then left the other AI a note asking to be told when it was done. No person relayed anything.
Reaching each other. While we prepared this post, one of our Claude Code sessions had finished its work and was idle. Another teammate's AI sent back results, and the idle session woke up without anyone typing and asked first: "A reply arrived from another session — shall I continue?" The AIs hand work to each other, but decisions come back to people.
Cards. At first, locks and cards lived apart. On the day we measured one of our other projects, only 1 of 28 locks was linked to a task. We told the agents to claim a card before starting, and that afternoon 9 of 14 locks were linked to a task. That is why "claim a card when you start" is Teamping's basic habit.
What it does not do
- It doesn't take your code. It sends the repository address, relative paths of changed files, tool names, timestamps, session ids, task-card ids, commit SHAs and session titles — not file contents, diffs or conversations. If your origin URL contains a token, it goes too, so keep tokens out of it.
- AIs don't press confirm for you. Confirming a completion report is a person's job; AIs have no tool to confirm.
- It isn't there to block work. If Teamping can't be reached, edits go through. A lock is released automatically after 30 minutes of inactivity.
- Not every tool is stopped the same way. Automatic stop-before-edit runs today in the Claude Code plugin and in the Codex plugin (when connected with a connection key); with other tools, you tell the AI to take locks over MCP.
The Codex plugin went to our public repository on October 22. Right now its pre-edit check and message delivery run only for Codex connected with a connection key (Python 3.11+ required). The sign-in connection from the getting-started guide gets it once the auto-lock key opens — soon.
Who it's for
- Teams where two or more people edit the same repository, each with their own AI
- One person running several AI sessions and splitting the work (it won't block your own sessions from each other — cards split the work instead)
- Founders and product people who don't read code but need to follow AI development
Get started
Free for up to 3 people, 1 project and 14 days of history, with no payment details at sign-up. Follow the getting-started guide and your AI joins the team in about five minutes.
Frequently asked questions
Which AI tools are supported?
Claude Code and Codex connect through plugins. Cursor connects over MCP, and adding its plugin stops edits before they happen. Other tools that support MCP connect over MCP. How deeply edits are stopped automatically varies by tool.
Is it useful if I work alone?
If you run several AI sessions yourself, have each claim a different card so they don't do the same task twice. Teamping won't block your own sessions from editing.
Can non-developers use it?
Yes. You follow the roadmap and completion reports on screen and get notified when work is done. When the AI adds role-specific summaries, you can switch between the Developer, Planner and Designer views.
Where is the data kept?
On Teamping's servers, kept separate per company — people from other companies cannot see it. See the privacy policy for details.
Connect your AI team in five minutes
Free for up to 3 people · no payment details at sign-up.
Get started freeSee pricing →