Skip to content

Running a Node

A node is the Spore client running on your machine. It downloads open-source models, serves them with llama.cpp, and earns credits for every community request it processes.

macOS (Homebrew):

Terminal window
brew install spore/tap/spore-node

Windows (Scoop):

Terminal window
scoop bucket add spore https://github.com/Chrisischris/scoop-bucket
scoop install spore-node

Start the node and follow the prompt. It opens the app so you can approve linking the device to your account. From then on the node authenticates with its own key and stays connected.

Manage everything from the Nodes page: pick a recommended model that fits your GPU, choose which context sizes to serve, and set the serving mode per model:

  • Personal: the model only answers your own requests. Free, private, and reachable from any of your devices.
  • Community: the model also processes requests from the network and earns credits.

New models run a short benchmark before serving community traffic, so the network knows what throughput to expect from your hardware.

  • You earn credits for the tokens your node processes on community requests.
  • A 5% platform fee comes out of community earnings.
  • Personal-mode requests are free. No credits move.

The network checks the token counts nodes report; nodes that consistently misreport get flagged and eventually excluded from job assignment.

To use Cursor or other OpenAI-compatible tools with the same end-to-end encryption as the web app, start the node’s localhost proxy (spore proxy start). See Encrypted Local Proxy.

On this page, turn on Automation runner to let the machine claim automation jobs. A node already running spore start picks that up. Model turns go through the encrypted local proxy on that machine (127.0.0.1:18790). The runner starts the proxy for the job and fails the job if the proxy cannot run. The prompt is not sent to the public API. The runner does not have to serve the model; inference can be on this machine or another. A host that should not run inference can use spore automation-runner instead. The switch is what Spore assigns from. Linked-node credentials from spore auth are enough; a separate sk_spore_ key is optional. GitHub is connected on Spore, not on the runner. See Automations. Interactive OpenCode is a separate setup.