Skip to content

XMTP AI

AI is transforming consumer tech, with messaging becoming the main channel for interacting with agent services. This shift will scale message traffic astronomically, analogous to the web’s rise in the 2000s. Just as Cloud-flare secured web traffic, messaging will need robust scalable end-to-end encrypted messages to protect sensitive communications.

Risks

Risks of not using end-to-end encryption:

Man in the Middle Attacks (MITM): Intercept requests in between to alter or manipulate data sent or received by the AI service or user.

  • Phishing: Messages can be intercepted and manipulated.
  • Privacy: Sensitive information read by unwanted parties.
  • Tampering: Content can be altered without detection.

Anonymity

Using ephemeral addresses can enhance security by allowing users to message agents anonymously, protecting their identities from exposure.

Backend

You can use the xmtp package to create wrapper around the messages sent and received by the AI agent.

Installation

Install the xmtp package

cmd
bun install xmtp

Usage

This is how you can use the xmtp package to create a client and handle messages.

import { XMTP } from "xmtp";
 
const xmtp = await XMTP(onMessage, {
  encryptionKey: process.env.WALLET_PRIVATE_KEY,
});
 
const onMessage = async (message, user) => {
  console.log(`Decoded message: ${message.content.text} by ${user.address}`);
  // Your AI model response
  await xmtp.sendMessage(response);
};

Gpt example

  1. handleMessage: Triggered each time a new message is received from XMTP.
  2. client.send(): Used to send messages (e.g., AI prompts and responses) back to the XMTP network.

In this example, when handleMessage is invoked, it takes the incoming user message, passes it to the AI model (OpenAI's Chat Completion API), and then sends the AI's response back over XMTP.

import { XMTP } from "xmtp";
import { OpenAI } from "openai";
 
const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});
 
const xmtp = await XMTP(onMessage, {
  encryptionKey: process.env.LOCAL_KEY,
});
 
const onMessage = async (message, user) => {
  console.log(`Decoded message: ${response} by ${user}`);
 
  // Prepare conversation history or context as needed
  const history = [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: message },
  ];
 
  // Call the OpenAI API to process the user's message
  const response = await openai.createChatCompletion({
    model: process.env.GPT_MODEL || "gpt-4",
    messages: history,
  });
 
  const response = response.data.choices[0].message.content.trim();
 
  // Send the AI-generated response back to the XMTP network
  xmtp.send(response);
};

Frontend

Same as the backend, you can use the xmtp-web package to create a client that decrypts and encrypts messages locally.

Installation

Install the xmtp-web package compatible with your frontend framework.

cmd
bun install xmtp-web

Usage

import { XMTP, Message } from "xmtp-web";
 
const xmtp = await XMTP(onMessage, {
  encryptionKey: process.env.LOCAL_KEY,
});
 
const onSend = async (text) => {
  return message;
};
 
const onMessage = async (message, agentAddress) => {
  console.log(`Decoded message: ${message} by ${agentAddress}`);
  let response; // Your ai model response
  await xmtp.sendMessage(response, agentAddress);
};

React example

This is how you can use the xmtp-web package to create a client and handle messages.

import { createClient, XMTP, Message } from "xmtp-web";
// ... other imports ...
 
function Chat({ user }: { user: UserInfo }) {
  const [messages, setMessages] = useState<Message[]>([]);
  const [newMessage, setNewMessage] = useState("");
  const [xmtp, setXmtp] = useState<XMTP | undefined>(undefined);
  const [isLoading, setIsLoading] = useState(true);
  // ... other state variables ...
 
  useEffect(() => {
    const init = async () => {
      const newWallet = Wallet.createRandom();
      // ... set wallet and recipient info ...
 
      try {
        if (user?.address) {
          await initXmtp(newWallet);
        }
      } catch (error) {
        console.error("Error resolving recipient:", error);
        setIsLoading(false);
      }
    };
 
    init();
  }, [user.address]);
 
  const onMessage = async (message: Message | undefined) => {
    if (message) {
      setMessages((prevMessages) => [...prevMessages, message]);
    }
  };
 
  const initXmtp = async (wallet: any) => {
    try {
      const xmtpClient = await createClient(onMessage, {
        privateKey: wallet.privateKey,
      });
      setXmtp(xmtpClient);
      setIsLoading(false);
    } catch (error) {
      console.error("Error initializing XMTP:", error);
      setIsLoading(false);
    }
  };
 
  const sendMessage = async (e: React.FormEvent) => {
    e.preventDefault();
    if (!xmtp || !newMessage || !recipientInfo?.address) return;
 
    try {
      const message = await xmtp.sendMessage(newMessage, user.address);
      setMessages((prevMessages) => [...prevMessages, message]);
      setNewMessage("");
    } catch (error) {
      console.error("Error sending message:", error);
    }
  };
 
  return <div className={styles.chatContainer}>{/* Render chat UI */}</div>;
}
 
export default React.memo(Chat);