Contributors share
Your machine runs an open model with llama.cpp when it is idle. One command installs a pinned, verified engine and keeps it running.
Invite-only beta · live on a community network
Share your GPU and RAM. Run AI. Earn compute for your own models.
$ noduma contribute --join ndm_j_…
✓ engine and model verified (SHA-256)
✓ node key stored, starts at login
● sharing qwen2.5 · coordinator app.noduma.dev
↳ answer streamed · credits +1
How it works
Your machine runs an open model with llama.cpp when it is idle. One command installs a pinned, verified engine and keeps it running.
Each request goes to a ready, fast contributor. If one fails, the answer moves to another. It never runs models and never stores prompts or answers.
Chat in the browser or point any OpenAI-compatible tool at Noduma. Researchers get affordable access to open models.
Built on trust you can check
Vetted reference contributors re-run a sample of answers and score them token by token. Only checks move a contributor's trust, and low trust suspends.
Credits come from what the coordinator measures, never from what a node reports.
Tokens arrive as they are produced; a failed node hands over without duplicating work.
The computer that answers can read your messages, so do not send secrets. Credits are not cashable yet. Today each model runs whole on one contributor; splitting one model across many machines is in the lab.
Contribute
The beta is invite-only. The operator sends you a join code.
macOS, Windows or Linux. Run noduma contribute --join <code>; it starts at login.
Every answer your machine serves earns credits to run models yourself.
Developers
Use your API key with /v1/models and streaming /v1/chat/completions at https://app.noduma.dev/v1. Works with the OpenAI SDKs and the tools built on them.
curl -N https://app.noduma.dev/v1/chat/completions \
-H "Authorization: Bearer $NODUMA_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "qwen2.5-1.5b-instruct-q4_k_m",
"stream": true,
"messages": [{"role": "user",
"content": "Hello"}]}'