Pooled

Pooled is a peer to peer system for running large open language models across several devices directly in the browser. A user starts a room, receives a short code, and types it on their other devices, such as a laptop, a desktop and a phone. The devices pool their memory, and the model is split across them by layer, so together they can run a model none of them could run alone. It is aimed at people who want to run big open models without renting servers or owning a single powerful machine. In the demo shown on the home page, three devices contribute 10 GB, 12 GB and 2 GB for a combined 24 GB, which is enough to run a 35 billion parameter mixture of experts Qwen model at about 21 tokens per second. Each device downloads only its own layers once and keeps them cached. Roughly 10 KB passes between devices per word. Anyone in the room can pick and start a model, and the pooled model can be used for group chat or for a coding agent that writes and edits files in a browser based workspace with a live preview and a console, including an option to auto approve edits. The site states the project is open source under the MIT license. No pricing page is published, and no paid plans are listed.