AI Signal 489
Show HN: Lumabri – What if LLMs worked like Napster?
Lumabri is a peer-to-peer system that runs mixture-of-experts language models across a swarm of machines, fetching weights on demand and caching them locally.
It allows engineers to pool disk and compute resources across standard CPUs without requiring upfront model downloads or GPU hardware. The system lowers the barrier to running large models by distributing the storage and execution load across anonymous peers.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Lumabri uses a pure C engine to run mixture-of-experts models across a peer-to-peer swarm without upfront downloads.
Weights are fetched on first use and cached locally, allowing subsequent queries to be served from disk even if the original server goes offline.
The system supports NAT traversal through a relay tracker and allows private swarms via an authentication token.
THE CLUSTER