Lumabri - Run huge MoE models on a P2P swarm

Show HN: Lumabri – Run Moe Models on a P2P Swarm with Colibri

Lumabri lets you run massive mixture-of-experts models like GLM, OLMoE, and DeepSeek on a peer-to-peer swarm. One machine shares a model, and any other machine can chat with it without downloading the entire model upfront—bytes are fetched on demand and cached locally. Built in pure C with no dependencies, it works on CPU and SSD first, with GPUs only making it faster. Anyone can join, even without a GPU, and donate disk or compute to help the swarm. It's a new way to democratize access to huge AI models.

Networks that pool GPUs recruit from the few; lumabri recruits from everyone.
  1. brainless

    I am sorry I did not understand all of it. But, would this allow running large MoE LLMs on a local network with experts spread out over multiple cheaper GPUs (or even CPUs)? This would perhaps be more useful than over the Internet, within offices for example.

  2. s2l

    What if one wishes to use various busybox nodes within the house?

    All the iot devices contributing to matmul but within a LAN?

  3. snovv_crash

    Cool idea. How do you handle temperature in the verification?

More from this day

2026-08-14