Every tool your team uses is multiplayer. Except AI.
Poly is multiplayer AI: a shared room where your team works with one AI agent, together. Everyone sees what was asked, anyone can answer or approve, and the decisions stay with the team.
Free to start. No card. Nothing to install.
You plan together. Then everyone prompts alone.
Your team is all-in on AI. But every session is private, so the work gets faster while the team gets further apart.
The context stays in private chats
What was asked, what was tried, what got rejected. It all builds up in four separate sessions that nobody else can see.
You become the middleman
You paste the output into Slack, explain how you got there, and answer questions about work the agent did.
The reasons disappear
The work gets merged. The understanding doesn't. A month later nobody remembers why the decision was made.
One room. One agent. The whole team.
Poly is a multiplayer AI coding workspace: your team and one agent, in one place, live.
Everyone sees the same thing
Every prompt, every answer and every change lands in one timeline, live. Nobody needs a recap, because nobody was out of the loop.
Anyone can answer, approve or steer
When the agent has a question or wants to do something big, it asks the room. Whoever knows the answer gives it. Nothing important happens without a yes.
The decisions stay with the team
Every call the team makes is kept with who made it and when. A new teammate reads it in thirty seconds. The agent remembers it next time.
Safe by default. The agent works in a sealed sandbox and asks before anything big. How security works
Every frontier lab. One room.
Claude, GPT, Grok and Muse, side by side. Each person picks their own. Switch mid-conversation and the room carries the context across.
- Claude Sonnet 5.5
- GPT GPT-6 Luna
- Grok Grok Build
- Muse Muse Spark 1.3
One free model from every lab. No API keys. Every lab on one plan, and each person picks their own model.
For the moments you'd normally screen-share.
Keep your solo sessions for solo work. Open a room when the work is shared.
George and Paul, PM and engineer
Speccing a feature
Notes go in, a spec comes out, and the engineer pushes back in the same room before anything is built.
Ringo and John, marketing and founder
Planning a launch
Brief, headlines and images in one place, with the founder answering the judgment calls as they come up.
John and Paul, founder and engineer
Building a prototype
One of you knows the customer, one of you knows the code. Both of you steer, and both of you watch it take shape.
Poly for Teams. Early access, a few teams at a time.
Your whole company, in rooms that know each other.
Bring the org in, sort people into teams, and give every team a room built for its work. What one team decides, the others' agents already know.
- Rooms per team Engineering, Marketing, Product, each with its own playbook
- Shared team memory Decisions and work the agent reads on every turn
- Cross-team rooms Ask Marketing's agent what Engineering shipped
- Usage by team One balance, receipts per person, caps per team
Design partners
Build Poly with us.
We're working closely with a small number of teams. Partners get a more curated Poly: one of us in your first rooms, a direct line to the founders, and the product shaped around the way your team actually works.
- A founder in your first roomsWe join, help, and get out of the way
- A direct line to usWeekly call, and a message away between calls
- Built around your workWhat you need next moves to the front of the line
- Partner pricingLocked in for as long as you stay
Stop prompting alone.
Open a room, invite one teammate, and do one real piece of work together.
Free to start. No card. Nothing to install. One link brings a teammate in.
Questions
What is multiplayer AI?
AI a group uses together: everyone works with the same AI agent in one shared place, sees what was asked and what it did, and can steer it, instead of each person prompting alone. That's Poly.
Is it really multiplayer, or just a shared history?
It's live. Everyone in the room sees prompts, answers and changes as they happen, and anyone can answer the agent's questions, approve what it does next, or stop it mid-flight.
Which models can we use?
Claude, GPT, Grok and Meta's Muse, in the same room. Each person picks their own model and thinking effort, and nobody needs an API key. Every lab has a free model — Sonnet 5.5, Haiku 5.5, GPT-6 Luna, Muse Spark 1.3 and Grok Build; any plan or top-up unlocks the rest — Opus 5.5, Fable 5.1, GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and Grok 4.7.
Do we have to stop using Claude Code or ChatGPT on our own?
No. Keep your solo sessions for solo work. Poly is for the work you'd do together anyway — the moments you'd normally screen-share.
Does it work with code we already have?
Yes. Connect a GitHub repository to a room: its work stays in the room, attributed to whoever made it, until you send it for review as a pull request or ship it. Anything meant only for the room can stay there. Or start from nothing.
What if the AI breaks something?
Every step is saved with a readable list of exactly what changed, and any step can be undone with the Undo button right under it. You can also require a yes before the agent edits anything at all.
What does it cost?
It's free to start, with no card. You pay only for the AI you run: plans from $20/month and top-ups from $5. Watching, chatting and approving in a room are always free.
Do you have a plan for teams?
Yes, in early access. Poly for Teams adds organizations, a room per team with its own playbook, shared team memory the agent reads on every turn, cross-team rooms, and usage by team. Book a demo or email support@usepoly.co.