Coordinate the swarm
Use gossip to coordinate peers across providers.
USE CASE / 02
Compute is everywhere.
Connect it.
Connect GPUs across clouds and homes for distributed training and inference.
“Doubling the network speed halves our compute budget. That’s the difference between a $1M model and a $500K model.”
Move the heavy parts
Transfer model data with blobs. Your app handles scheduling.
Keep another path open
Relay fallback keeps otherwise unreachable peers connected.
IN PRACTICE
Built with iroh.
Use case · Distributed training
Nous Research / Psyche
Psyche brings GPUs across clouds and homes together to train models. Nous Research uses iroh gossip to coordinate the peers and blobs to move training data.
Explore Nous Research↗
Use case · Shared inference
Block / Buzz
Buzz brings people and agents into one workspace. MeshLLM is built into Buzz, letting community members opt in to share spare compute and run open models together over iroh.
Read the Buzz story↗
“With iroh, every node is just a public key. We pool GPUs across clouds and closets into one mesh. No central server, no NAT config.”
Michael NealePrincipal Engineer at Block
Need a hand with the network?
You build the AI. We can run your relays and help tune your deployment for performance and reliability.