The ICP Agent Kit includes a powerful LangChain integration that enables natural language interaction with the Internet Computer blockchain. This integration provides pre-built tools, specialized agents, and natural language processing capabilities.
Prerequisites: You’ll need an OpenAI API key to use the natural language features. Set it as an environment variable (OPENAI_API_KEY) or pass it in the configuration.

​
Architecture Overview

LangChain Architecture The LangChain integration consists of three main components:

​
1. Natural Language Processor

The NLP processor uses OpenAI’s GPT-4 to understand natural language commands and map them to the appropriate tools.

​
2. Tool System

10 pre-built LangChain tools that wrap the core plugin functionality:
  • Identity Tools: Create and manage identities
  • Token Tools: Transfer tokens and check balances
  • Canister Tools: Deploy and manage canisters
  • Cycles Tools: Monitor and top-up cycles

​
3. Specialized Agents

Purpose-built agents for different use cases:
  • General Agent: All-purpose blockchain operations
  • Developer Agent: Focused on canister development
  • DeFi Agent: Optimized for token operations
  • Governance Agent: NNS and DAO operations

​
Quick Start

​
Basic Setup

import { ICPAgent } from 'icp-agent-kit';

// Initialize with OpenAI
const agent = new ICPAgent({
  network: 'mainnet',
  openai: {
    apiKey: process.env.OPENAI_API_KEY,
    model: 'gpt-4-turbo-preview', // optional, this is the default
  },
});

await agent.initialize();

​
Natural Language Commands

Once initialized, you can use natural language to interact with the blockchain: 
// Direct natural language processing
const result = await agent.processNaturalLanguage('Transfer 5 ICP to ryjl3-tyaaa-aaaaa-aaaba-cai');

// Check balance
const balance = await agent.processNaturalLanguage('What is my ICP balance?');

// Deploy a canister
const deployment = await agent.processNaturalLanguage(
  'Create a new canister with 2 trillion cycles',
);

​
Using Specialized Agents

For more focused interactions, use specialized agents: 
// Create a DeFi agent for token operations
const defiAgent = agent.createAgent('defi');
const response = await defiAgent.chat('I want to transfer some tokens. What do I need to know?');

// Create a developer agent for canister work
const devAgent = agent.createAgent('developer');
const help = await devAgent.chat('How do I deploy a WASM module to my canister?');

​
Key Features

Context Awareness

Agents maintain conversation context, understanding references to previous operations.

Error Handling

Natural language error messages help users understand and fix issues.

Multi-Step Operations

Complex workflows can be expressed in simple commands.

Tool Transparency

See exactly which tools are being used for each operation.

​
Available Commands

Here are some example commands you can use:

​
Identity Management

  • “Show me my current identity”
  • “Create a new identity called alice”
  • “Switch to my trading identity”
  • “Generate a new seed phrase”

​
Token Operations

  • “Check my ICP balance”
  • “Transfer 10 ICP to bob.icp”
  • “Show me all my token balances”
  • “Transfer 1000 CHAT tokens to principal abc123”

​
Canister Management

  • “Create a new canister with 5T cycles”
  • “Deploy my WASM to canister xyz”
  • “Check the status of all my canisters”
  • “Add alice as a controller to my canister”

​
Cycles Management

  • “Check cycles balance for canister abc”
  • “Top up my canister with 2T cycles”
  • “How many cycles can I get with 1 ICP?”
  • “Monitor my canister cycles and alert me when low”

​
Configuration Options

​
Model Selection

You can choose different OpenAI models based on your needs: 
const agent = new ICPAgent({
  network: 'mainnet',
  openai: {
    apiKey: 'your-api-key',
    model: 'gpt-4-turbo-preview', // Best performance
    // model: 'gpt-3.5-turbo' // Faster and cheaper
  },
});

​
Environment Variables

The integration supports these environment variables:
  • OPENAI_API_KEY: Your OpenAI API key
  • ICP_NETWORK: Default network (mainnet/local)

​
Best Practices

​
What’s Next?

Explore Tools

Learn about all 10 available LangChain tools

Natural Language Guide

Master natural language commands

Specialized Agents

Deep dive into each agent type

Examples

See real-world examples