// POOLED P2P MESH & MODEL CONTROLS
pooled

Room settings

1 device (1)

Joining

Ask before new devices joinA device that types the room code waits until you allow it. Your invite link and QR code let a device in without asking.

Conversation

Who sees the chat
Answer length
Think before answeringSlower, more careful: the model reasons before it answers.

Model

Answer styleA voice for every answer, from the next question on.
Wording
Layer split
Re-deal on its own experimentalWhen a device drops and is not back within a minute, deal its layers to the devices still here; an answer or Code run in progress carries on.

Room

DocsHow rooms, the Serve API and Code mode work (new tab) Clear cached weightsFrees this device's disk; the next start downloads again

    A device wants to join

    It typed the room code. Once in, it can hold layers and see the chat.

    Start a room, or join one.

    Each device in a room shares some memory, and together they run one open AI model in your browser tabs.

    MemoryGBChat only, no memory to give

    New room

    You get a code to type on your other devices.

    Have a code?

    Older rooms have four. You can also paste an invite link.

    Joining the room

    Details

    Runs in Chrome or Edge, or Safari 26, on a laptop or desktop. A phone alone can only chat. The smallest model is at most a 1.8 GB download.

    Network

    Can't connect from this network? A relay (TURN) server carries the traffic when two devices can't reach each other directly (work networks that block UDP). When this site has one set up, rooms use it by themselves; here you can set your own instead (coturn, for example, or a provider's). Saved in this browser only; join links never include it.

    Room

    Ready
    tok/s
    -
    ms per word
    -
    tokens
    0
    context
    -
    Pick a model and press Start

    Pick a model

    Anyone in the room can start it.
    MODEL

    Each device downloads only its own layers, once. They stay cached for next time.

    No device in this room can run a model: none has WebGPU. Join from a laptop or desktop in Chrome or Edge (or Safari 26) to start one.

    The room pools0 GB

      This device gives

      GB

      A pledge is a promise: this device never holds more. It covers the layers it holds and their context memory, and on the device that hosts the model also the embedding and output head.

      This browser can't use its graphics chip, so this device can chat but not help run the model.

      Invite more devices

      ----

      Scan it, or type the code on another device. Each device adds the memory it gives.

      Loading

      Each device downloads only its own layers.

      Project

      Files

      No files yet

      Agent

      No port servedsandbox
      Your app shows up hereThe agent serves it on a port like :5173
      0 logs
      
          
      Open a file from the list to read or change it.
      Saving updates the project: the preview reloads and the agent sees the change.
      
          
      
        
      
      
      
      
      
      
        
        
        
      
      
    
      
  
      

      


      
  
      
    
    
      Invite a device
      
    
      Open the link or scan the code on another device. It joins this room and adds its memory.
      
    
      
    
      On your phone? Scan with the camera app.
      
    
      or type the room code
      
    
      
    
      
    
      
  
      

      

      
  
      
    
    
      Room card
      
    
      A picture of this room's devices and speed, to keep or share.
      
    
    
      
  
      

      

      
  
      
    
    
      Room over
      
    
      
    
      
  
      

      

      
  
      
    
      Leave the room?
      
    
      
    
      
  
      

      

      
  
      
    pooled
    Room ----· 1 device
    
  
      
  
      
    
      
      
      
      
      
        
      Standing by
      
        
      hop0 tok/s0 tokens here- per hop
      
        
      
        
      
      
      
      
      
        
        
        
        
        Keep this screen on, plugged in and in front.
        
      
        
    
        
    
        
      
        
        
        Serve API
        
        
        Use this room from your own tools: an OpenAI and Anthropic endpoint on your computer.
        
        
        Waiting for a client to connect
        
        
        Run step 1. Your tool shows up here when it connects.
        
        
        
      
        
      
        
        
        Connected now
        
        
          
      
          
      
          
        How to connect
        
          
          
          1Step 1: Run this on your computer
          
          
          $
          
        
          
        
          
          
          2Step 2: Point any OpenAI or Anthropic tool here
          
          
          
            
          OpenAIhttp://127.0.0.1:8080/v1
          
            
          Anthropichttp://127.0.0.1:8080
          
          
          
          
          Any API key works. In Continue or Open WebUI (Settings, Connections), add an OpenAI connection with the OpenAI URL. Used --port? Change 8080.
          
        
          
        
          
          
          3Step 3: Try it
          
          
          
            
          
              
          
                
                
                
                
              
          
              
            
          
            
          
            
          
          
          
        
          
      
          
      
          
        
        
          Needs Node 22 or newer. Requests use this room's GPUs and show in the chat under the room's visibility setting.