CA:9FpoqFaowv1JNJq9xXuQxoXcSnKTWJD1J2xVVe2Zpump

Documentation

Technical reference for the NodevaLM protocol

Introduction

NodevaLM is a decentralized AI inference protocol built on Solana. It enables developers to deploy, run, and monetize large language models with zero-knowledge privacy guarantees.

The protocol consists of three core layers:
- Inference Layer: Distributed GPU nodes running NVIDIA NIM
- Settlement Layer: Solana smart contracts for payments and verification
- Privacy Layer: ZK circuits for private computation proofs

Quick Start

Install the NodevaLM SDK:

$ npm install @nodevalm/sdk

Initialize a client:

import { NodevaLM } from '@nodevalm/sdk';

const client = new NodevaLM({
  network: 'mainnet-beta',
  wallet: yourSolanaWallet
});

const response = await client.inference({
  model: 'nodevalm-7b',
  prompt: 'Explain zero-knowledge proofs',
  maxTokens: 256
});

Architecture

NodevaLM uses a modular architecture designed for horizontal scaling:

INFERENCE ROUTER
Routes requests to optimal nodes based on latency, load, and cost. Implements weighted round-robin with health checks.

NODE REGISTRY
On-chain registry of active inference nodes with staking requirements and performance metrics.

PAYMENT STREAM
Implements x402 protocol for micropayments. Supports both prepaid credits and real-time streaming.

ZK VERIFIER
Verifies computation proofs on-chain. Uses Groth16 for efficient verification.

API Reference

POST /v1/inference
Submit an inference request.

Request body:
{
  "model": "nodevalm-7b",
  "prompt": "string",
  "max_tokens": 256,
  "temperature": 0.7,
  "private": false
}

Response:
{
  "id": "inf_abc123",
  "model": "nodevalm-7b",
  "output": "string",
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 256
  },
  "proof": "0x..." // if private=true
}

ZK Privacy

NodevaLM supports private inference using zero-knowledge proofs. When enabled, your prompts and outputs remain encrypted while still being verifiable.

Enable private mode:

const response = await client.inference({
  model: 'nodevalm-7b',
  prompt: 'sensitive query',
  private: true
});

// Response includes ZK proof
console.log(response.proof);

// Verify proof on-chain
await client.verifyProof(response.proof);

Proof generation adds ~2.4 seconds overhead but provides cryptographic privacy guarantees.