@monad-agent-kit/ai-langchain

LangChain integration for Monad Agent Kit — converts all registered agent actions into LangChain tools.

Installation

bun add @monad-agent-kit/core @monad-agent-kit/ai-langchain @langchain/core

Usage

import { MonadAgentKit } from "@monad-agent-kit/core"
import { tokenPlugin } from "@monad-agent-kit/plugin-token"
import { defiPlugin } from "@monad-agent-kit/plugin-defi"
import { createLangChainTools } from "@monad-agent-kit/ai-langchain"
import { ChatOpenAI } from "@langchain/openai"
import { AgentExecutor, createToolCallingAgent } from "langchain/agents"
import { ChatPromptTemplate } from "@langchain/core/prompts"

// 1. Set up the agent with plugins
const agent = new MonadAgentKit({ privateKey: process.env.PRIVATE_KEY })
await agent.use(tokenPlugin)
await agent.use(defiPlugin)

// 2. Convert all actions to LangChain tools
const tools = createLangChainTools(agent)

// 3. Use with a LangChain agent
const llm = new ChatOpenAI({ model: "gpt-4" })
const prompt = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful blockchain assistant."],
  ["human", "{input}"],
  ["placeholder", "{agent_scratchpad}"],
])

const langchainAgent = createToolCallingAgent({ llm, tools, prompt })
const executor = AgentExecutor.fromAgentAndTools({ agent: langchainAgent, tools })

const result = await executor.invoke({
  input: "What is my MON balance?",
})

How It Works

createLangChainTools() takes a MonadAgentKit instance and returns an array of DynamicStructuredTool objects:

• Action name becomes the tool name
• Action description becomes the tool description
• Action parameters (JSON Schema) are converted to Zod schemas
• Tool execution calls agent.execute() and returns JSON-stringified results

Any plugins registered before calling createLangChainTools() will have their actions included.

Peer Dependencies

• @langchain/core >= 0.3.0

License

MIT