Host a node

Hosting is how the network exists. When you run diffuse host, your machine holds a slice of a model and serves compute to clients. This is contribution, not consumption.

diffuse host --model Qwen/Qwen2.5-0.5B-Instruct

Your node profiles its hardware, takes the slice the network needs most, announces it over gossip, and starts serving. It runs in the foreground so you can watch the logs. Ctrl+C stops it cleanly and shuts down its worker.

Options

Flag	Default	Meaning
--model <id>	required	the model to serve a slice of
--worker <url>	http://127.0.0.1:50051	the worker endpoint
--listen <addr>	0.0.0.0:9440	gossip listen address
--bootstrap <urls>	built-in sentinels	sentinels to join through
--overhead <f>	0.3	memory fraction held back as headroom
--public-addr <addr>	auto	the address to advertise to peers

Behind a NAT or firewall

Diffuse handles it. On startup your node checks whether it can accept inbound connections. If it cannot, it automatically serves its compute through a sentinel relay, so it still contributes rather than only consuming. The relayed compute stays encrypted end to end; the relay sees only ciphertext.

Keep it running

To keep contributing after you close the terminal, run it detached:

nohup diffuse host --model Qwen/Qwen2.5-0.5B-Instruct > ~/.diffuse/host.log 2>&1 &

Watch it with tail -f ~/.diffuse/host.log, stop it with pkill diffuse.

How much you need

You do not need to hold a whole model. A node holding three layers of a 64-layer model is a useful contribution. Diffuse measures each model's real memory cost and hands you a slice that fits. A small machine takes a small slice; that is the design, not a limitation. See choosing a model.

Next

• Choosing a model
• Replication and healing