nilai-ai-sdk-provider

A custom Vercel AI SDK provider for Nillion's nilAI - privacy-preserving AI that runs in a trusted execution environment (TEE).

Prerequisites

To use nilAI, you'll need:

1. Nillion Wallet: https://docs.nillion.com/community/guides/nillion-wallet
2. Testnet NIL tokens: https://docs.nillion.com/community/guides/testnet-faucet
3. nilAI API Key: https://docs.nillion.com/build/network-api-access

Installation

npm install nilai-ai-sdk-provider ai dotenv

Quick Start

1. Create a .env file:

NILLION_NILAI_API_KEY="your-api-key-here"

2. Use the provider:

import { generateText } from 'ai';
import { nilai } from 'nilai-ai-sdk-provider';
import 'dotenv/config';

const result = await generateText({
  model: nilai(),
  prompt: 'Explain quantum computing in simple terms',
});

console.log(result.text);

Examples

See /examples for complete examples:

• Basic chat
• Streaming
• Web search

Configuration

Provider Settings

const model = nilai({
  temperature: 0.7, // 0-1, default 0.7
  maxTokens: 1000, // default 1000
  topP: 0.9, // 0-1, default 0.9
  webSearch: true, // Enable web search
});

Per-request Options

const result = await generateText({
  model: nilai(),
  prompt: 'Latest AI news',
  providerOptions: {
    nilai: { webSearch: true },
  },
});

Features

• Privacy-preserving: All inference runs in a TEE
• Streaming support: Real-time response streaming
• Web search: Automatic date context injection for temporal queries
• TypeScript native: Full type definitions included
• Error handling: Comprehensive error messages

Rate Limits

Testnet limits:

• 500 requests/day
• 100 requests/hour
• 10 requests/minute
• 100 web searches

API Reference

nilai(settings?)

Creates a language model instance.

createNilai(options)

Advanced configuration with custom base URL and headers.

See TypeScript definitions for complete API documentation.

Support

• 📖 Documentation
• 🐛 Issues
• 💬 Discord

License

MIT - see LICENSE