Install

One command turns any Mac or Linux box into a working provider node. No clone, no Docker, no git.

Install a provider node

bash
curl -fsSL https://gpunetwork.xyz/install | sh

The installer is a single bash script served by the orchestrator. It:

  • Verifies you have python3
  • Installs Ollama if missing (via Homebrew on macOS or the official installer on Linux)
  • Starts the Ollama daemon and pulls a default model (~2GB)
  • Writes ~/.gpu/provider.py from the orchestrator (no separate download)
  • Creates a Python venv and installs the one dependency (httpx)
  • Registers the node with the orchestrator and starts serving inference

Configuration

Optional config file lives at ~/.gpu/config.toml. The installer doesn't create it — drop one in to customise:

toml
[node]
name = "my-rig"
backend = "ollama"
model = "llama3.2:3b"
prevent_sleep = true

[wallet]
public_address = "YOUR_SOLANA_ADDRESS"

[network]
orchestrator_url = "https://gpunetwork.xyz"

Run the node by hand

If you cloned the repo and want fine-grained control:

bash
python3 -m venv .venv
source .venv/bin/activate
pip install httpx
python provider/provider.py --orchestrator https://gpunetwork.xyz --wallet <pubkey>

Browser node (no install)

Open https://gpunetwork.xyz/provide in Chrome/Brave/Edge/Arc. Pick a model size and click Start providing. The model loads via WebGPU into your tab — no terminal needed. Stop by closing the tab.

Tip
Browser nodes are easiest to onboard but pay less per task (smaller models, lower throughput). For best earnings run a native node with Ollama on a real GPU.