@layr-labs/agentkit-opacity

Opacity adapter for verifiable inference in EigenLayer AgentKit. This adapter integrates with Cloudflare's AI Gateway and Opacity's zkTLS proof system to provide verifiable AI inference capabilities.

Installation

npm install @layr-labs/agentkit-opacity
# or
yarn add @layr-labs/agentkit-opacity
# or
pnpm add @layr-labs/agentkit-opacity

Features

• 🔒 Verifiable AI inference using Opacity's zkTLS proofs
• 🤖 Support for OpenAI and Anthropic models through Cloudflare AI Gateway
• ✅ Proof generation and verification
• 🛠️ Highly configurable model parameters
• 💬 Full chat conversation support
• 🎯 Advanced sampling options (top-p, top-k)

Usage

import { OpacityAdapter, ModelProvider, ChatMessage } from '@layr-labs/agentkit-opacity';

// Initialize the adapter
const adapter = new OpacityAdapter({
  teamId: process.env.OPACITY_TEAM_ID!,
  teamName: process.env.OPACITY_TEAM_NAME!,
  apiKey: process.env.OPACITY_API_KEY!,
  opacityProverUrl: process.env.OPACITY_PROVER_URL!,
  modelProvider: ModelProvider.ANTHROPIC, // or ModelProvider.OPENAI
});

// Simple text generation with a single prompt
const simpleResult = await adapter.generateText('What is the meaning of life?', {
  model: 'gpt-4oo',
  temperature: 0.7,
  maxTokens: 1000,
});

// Chat conversation with multiple messages
const conversation: ChatMessage[] = [
  {
    role: 'system',
    content: 'You are a helpful AI assistant.'
  },
  {
    role: 'user',
    content: 'What is quantum computing?'
  },
  {
    role: 'assistant',
    content: 'Quantum computing is a type of computing that uses quantum phenomena...'
  },
  {
    role: 'user',
    content: 'Can you explain quantum entanglement?'
  }
];

const chatResult = await adapter.generateText(conversation, {
  model: 'claude-3-5-sonnet-20241022',
  temperature: 0.7,
  maxTokens: 4096,
  topP: 0.9,
});

console.log('Generated text:', chatResult.content);
console.log('Proof:', chatResult.proof);

// Verify the proof
const isValid = await adapter.verifyProof(chatResult.proof);
console.log('Proof is valid:', isValid);

Configuration

Environment Variables

OPACITY_TEAM_ID=your_team_id
OPACITY_TEAM_NAME=your_team_name
OPACITY_API_KEY=your_api_key
OPACITY_PROVER_URL=https://your-opacity-prover-url

Adapter Options

The adapter supports various configuration options:

interface OpacityAdapterConfig {
  teamId: string;
  teamName: string;
  apiKey: string;
  modelProvider?: ModelProvider; // OPENAI or ANTHROPIC
  opacityProverUrl: string;
  gatewayUrl?: string;
}

Chat Messages

You can pass a conversation history using the ChatMessage interface:

interface ChatMessage {
  role: 'system' | 'user' | 'assistant';
  content: string;
}

Model Options

When calling generateText, you can customize the model behavior:

interface GenerateTextOptions {
  model?: string;                // Model name (e.g., 'gpt-4oo', 'claude-3-5-sonnet-20241022')
  temperature?: number;          // Controls randomness (0-1)
  maxTokens?: number;           // Maximum tokens to generate
  systemPrompt?: string;        // System prompt for single-prompt generation
  topP?: number;                // Nucleus sampling parameter (0-1)
  topK?: number;                // Top-k sampling parameter
}

Supported Models

OpenAI Models

• gpt-4oo
• gpt-4o-mini

Anthropic Models

• claude-3-5-sonnet-20241022
• claude-3-opus-20240229
• claude-3-sonnet-20240229
• claude-3-5-haiku-20241022

Each model comes with optimized default parameters that can be overridden using GenerateTextOptions.

Error Handling

The adapter throws the following error types:

• ProofGenerationError: When proof generation fails
• ProofVerificationError: When proof verification fails

Contributing

Please read the contributing guidelines in the root of the monorepo for details on our code of conduct and the process for submitting pull requests.

License

MIT License - see the LICENSE file for details.