Pooled

Pooled runs a single open model across laptops, desktops and phones in browser tabs. It lets people combine their devices' memory to use a model that wouldn't fit on one device, for shared chat or coding tasks.

• Pooled GPU inference: Participating devices run assigned model layers through WebGPU and communicate over WebRTC; inference happens on the room's devices.
• Model selection: Choose Qwen3 1.7B, Qwen 3.8 27B or Qwen 3.6 35B MoE, with each device downloading only its assigned layers and caching them for later use.
• Browser coding agent: Code mode writes and edits small apps, displays a sandboxed preview, reads console errors and attempts fixes; room members can view the files and preview.
• Connections to other tools: A local API bridge supports OpenAI Chat Completions, OpenAI Responses and Anthropic Messages, including tool calling, for clients such as Claude Code, Codex CLI, Continue and Open WebUI.

Pooled is open source under MIT and requires no account or installation on participating devices. The app runs in the browser and can also be served locally. Devices need WebGPU-capable GPUs and enough combined memory for the chosen model. Model downloads come from Hugging Face or another room member.

Code files stay in the host's browser or a chosen disk folder. Changes to a real folder require the host's approval.

Rooms share conversation content with their members, subject to the host's visibility settings. Intermediate model data isn't private against a determined peer, so use rooms with people you trust. Where a site provides a relay, connections can fall back to it when direct links are blocked.