Quickstart

Once Diffuse is installed, you are three commands away from talking to the network.

Talk to the network

diffuse chat

diffuse chat joins the network through the built-in sentinels, finds a servable model, starts a local tokenizer, and streams the answer back token by token. Type your message, or /quit to leave.

See what is live

diffuse models

This lists the models the network can serve right now, meaning every layer is covered by at least one reachable node. A model that is only partially held is marked incomplete and is not listed as servable.

Ask one question

diffuse query --prompt "Explain black holes in two sentences."

query is the one-shot form: it discovers a model, routes through it, prints the answer, and exits.

Contribute your machine

Give the network capacity by hosting a slice:

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

Your node profiles its own hardware, takes the slice the network needs most, announces it, and starts serving. Leave it running to keep contributing.

Use it from any OpenAI client

Run the local server and point any OpenAI-compatible tool at it:

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

curl http://localhost:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"Qwen/Qwen2.5-0.5B-Instruct","messages":[{"role":"user","content":"Hello"}]}'

See the OpenAI server guide to connect LibreChat, Open WebUI, or Continue.

Where to go next

• Chat with the network
• Host a node
• Choosing a model
• CLI reference