Show HN: Lumabri – What if LLMs worked like Napster?

Hacker News reports on Lumabri, an experimental project that distributes large language model inference across networks of ordinary computers. Rather than requiring a single powerful machine, it treats networked computers as a shared resource pool—one provides storage, another compute, another different model sections. The approach suits Mixture-of-Experts models, which have hundreds of billions of parameters but activate only a fraction per request; instead of moving massive expert weights, the system sends small activations to peers already holding those experts. The creator, who previously built Colibrì to run LLMs on single machines, is still working through challenges in network latency, security, and reliability. The goal is for everyday computers to contribute spare resources while the peer-to-peer swarm handles the rest—a vision inspired by networks like Napster, where users become the infrastructure.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton