Chat
Ask the room’s model anything. See Chat.
Getting started
What is Pooled
Pooled runs one open model across the devices in a room, in browser tabs. Nothing to install, no account, and no server does the thinking.
Pooled runs one AI model across several devices at once. Your laptop, a friend’s desktop and a phone each hold part of the model and run it on their own GPU, in a browser tab. Together they run a model none of them could run alone.
There is nothing to install and no account. The model runs on your devices, not on a server.
How it works in 30 seconds
Section titled “How it works in 30 seconds”- One device opens pooled.run/room and starts a room. It gets a code like
4TK-G9P. - Other devices join with the code, a link or a QR code. Each one gives the room some memory.
- Someone picks a model and presses Start the model. Each device downloads only its part.
- Everyone in the room can ask the model questions.
One laptop is enough for the smallest model. More devices let you run bigger ones:
| Model | Memory the room needs |
|---|---|
| Qwen3 1.7B | 5.6 GB at a 16K context, 4 GB at 8K |
| Qwen3.8 27B | 17 GB |
| Qwen3.6 35B MoE | 22.5 GB |
What you can do with a room
Section titled “What you can do with a room”Code mode
Ask for a small app. The model writes it and runs it in a preview. See Code mode.
Serve API
Use the room from Codex CLI, Claude Code, opencode or the OpenAI and Anthropic SDKs. See Serve API.
Terminal
Join or host a room from the command line. See Terminal.
Start here
Section titled “Start here”Start a roomRun your first model on one laptop.
Invite devicesAdd a phone, a second laptop or a friend's desktop.
Want a step-by-step walk-through? The starter kits take you through each way to use Pooled: one laptop, two devices, Code mode, coding tools, the terminal and OpenClaw.
Curious how it works under the hood? See Architecture.