Skip to content
Brian

XMTP Integration

All the agents inside the @brian-ai/langchain package can leverage an XMTP HandlerContext as a callback handler to send the agent's thoughts and actions in an XMTP chat.

What are callback handlers?

Langchain offers a powerful callback system that allows you to hook into the various stages of your LLM application. This is useful for logging, monitoring, streaming, and other tasks.

XMTPCallbackHandler

We developed a powerful XMTPCallbackHandler that allows you to send your agent's thoughts and actions to an XMTP chat.

It's also fully customizable and you can choose which callbacks are handled and which are not via its configuration:

type XMTPCallbackHandlerOptions = {
  onChainStart?: boolean; // sends a message when the chain starts
  onChainEnd?: boolean; // sends a message when the chain ends
  onChainError?: boolean; // sends a message when the chain errors
  onLLMStart?: boolean; // sends a message when the LLM starts
  onLLMEnd?: boolean; // sends a message when the LLM ends
  onLLMError?: boolean; // sends a message when the LLM errors
  onToolStart?: boolean; // sends a message when a tool starts
  onToolEnd?: boolean; // sends a message when a tool ends
  onToolError?: boolean; // sends a message when a tool errors
  onRetrieverStart?: boolean; // sends a message when the retriever starts
  onRetrieverEnd?: boolean; // sends a message when the retriever ends
  onRetrieverError?: boolean; // sends a message when the retriever errors
  onAgentAction?: boolean; // sends a message when the agent performs an action
};

Usage

In the context of an XMTP bot, you just need to pass the HandlerContext to your agent and define the callbacks options:

import { HandlerContext, run } from "@xmtp/message-kit";
import { createBrianAgent } from "@brian-ai/langchain";
import { ChatOpenAI } from "@langchain/openai";
 
run(async (context: HandlerContext) => {
  const brianAgent = await createBrianAgent({
    apiKey: process.env.BRIAN_API_KEY!,
    privateKeyOrAccount: process.env.AGENT_PRIVATE_KEY as `0x${string}`,
    llm: new ChatOpenAI({ apiKey: process.env.OPENAI_API_KEY }),
    xmtpHandler: context,
    xmtpHandlerOptions: {
      onAgentAction: true,
      onToolStart: true,
    },
  });
  const { content } = context.message;
  const result = await brianAgent.invoke({ input: content.text });
  await context.send(result.output);
});

You can also use a CDP wallet agent with the XMTP handler:

import { HandlerContext, run } from "@xmtp/message-kit";
import { createBrianCDPAgent } from "@brian-ai/langchain";
import { ChatOpenAI } from "@langchain/openai";
 
run(async (context: HandlerContext) => {
  const brianAgent = await createBrianCDPAgent({
    apiKey: process.env.BRIAN_API_KEY!,
    coinbaseApiKeyName: process.env.COINBASE_API_KEY_NAME!,
    coinbaseApiKeySecret: process.env.COINBASE_API_KEY_SECRET!,
    llm: new ChatOpenAI({ apiKey: process.env.OPENAI_API_KEY }),
    walletData: JSON.parse(process.env.COINBASE_WALLET_DATA!),
    xmtpHandler: context,
    xmtpHandlerOptions: {
      onAgentAction: true,
      onToolStart: true,
    },
  });
  const { content } = context.message;
  const result = await brianAgent.invoke({ input: content.text });
  await context.send(result.output);
});