Poly - Multiplayer Claude Code for you and your friends

Hey Product Hunt 👋

I'm Richard — my cofounder Mitchell and I built Poly because of a problem we had ourselves: AI coding is single-player.

When one of us ran a Claude Code session, the other saw nothing. Not the prompt, not the diff, not the cost. Work that used to happen in pull requests — visible, reviewable, attributable — was happening invisibly in a terminal. The better the agent got, the more of our codebase was being written somewhere neither of us could see.

So we built the room we wanted:

One shared timeline. Every prompt, agent reply, tool call, diff, and turn receipt lands in front of the whole team, live.

The agent asks before it edits, and any teammate can approve — in our demo video, one person sends the prompt and the other approves the edit from their own screen. Any turn reverts with one click. Four working modes, from full-auto to approve-everything.

Every turn has a receipt. Who ran it, which model, how long it took, what it cost — down to the cent, visible to the whole room. No mystery meters.

Shared mode is everyone steering one agent in turns. Multi-agent mode gives each person their own agent on its own git branch, merged when you're ready. Every turn runs in an isolated sandbox with allowlist-only network access.

And there's a group chat the agent never sees — for arguing about the prompt before you send it.

We built all of Poly inside Poly — every feature was written by agents in its own rooms with both of us watching.

Poly is currently open beta, free, no API keys or setup — sign up and you're in a room in under a minute. Desktop only for now.

We're in the comments all day — ask us anything. We'd rather answer it straight than have you wonder.