Getting started

Quickstart

Three ways in: use the network, verify the claims yourself, or contribute compute and earn.

Call the API

The inference endpoint is OpenAI-compatible, so most existing SDKs and tools work by changing the base URL. Prompts are billed in tokens and are not attached to a profile.

curl https://api.nuroai.xyz/v1/chat/completions \
  -H "Authorization: Bearer $NURO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nuro-max-27b",
    "messages": [{ "role": "user", "content": "Explain private inference." }]
  }'

Models

Native GPU workers advertise larger models (for example nuro-max-27b); in-browser WebGPU workers advertise lighter ones (for example nuro-browser-8b). Availability depends on who is online in the swarm.

The paper

The privacy and correctness claims are not asks for trust. The full write-up — adversary, defenses, receipt, and measured gates — is the Paper on the site.

nuroai.xyz/paper

What the gates show

Gate A: the split run is bit-identical to the whole model. Gate D: obfuscation drops recovery to chance on proprietary weights, and fails on open weights. Gate E: one MPC share recovers the prompt at chance even with public weights.

Contribute a GPU

A worker is a small client that connects to the orchestrator over a WebSocket and serves inference jobs. There are two ways to run one:

Native. Point the worker at a local Ollama install with a capable model. Best throughput, best earnings.
Browser. Open the /earn page on a WebGPU-capable browser and it becomes a worker with zero install.

See Workers and earning for setup, model classes, and how payouts work.

Stake $NURO

If you would rather back the network than run it, stake $NURO from self-custody and earn a share of real network revenue. Only your wallet can unstake or claim. See Staking and treasury.