Build an AI-Powered Email Agent with Reusable Code

Why Automate Your Tasks?

Hi all ?! As an app developer, I’m excited to share how you can create simple yet powerful agents to automate your daily tasks.

? Like many of you, I receive an overwhelming number of emails every day. Despite my best efforts, achieving the elusive Inbox Zero remains a challenge. Sorting through emails like order confirmations and shipping updates is tedious and time-consuming.

But here’s the good news: automation can save the day!

? I’ve written a basic script leveraging AI to help automate email categorization—and you can, too.

In this article, I’ll share reusable code snippets to help you build your own automation agents tailored to your needs. ?

---

The Beauty of Automation

There are countless tools, including no-code platforms, that can handle entire processes for you. However, I prefer breaking tasks into modular code snippets. Why?

1. Flexibility: Modular code can easily adapt to changing requirements.
2. Reusability: Write once, use forever.
3. Maintainability: Small, independent chunks of code are easier to debug and improve.

By taking an incremental approach, you can gradually replace manual steps with automated ones.

?‍? My go-to tool for prototyping scripts is Znote—a notebook with live coding and AI that helps me track and enhance my workflows. Give it a try, or use your favorite IDE!

---

Let’s Build an Email Automation Agent

Goals

When a new email arrives, we want to:

1. Read the email and extract relevant information.
2. Use an LLM to determine its category from a predefined list.
3. Create Gmail labels (if they don’t already exist).
4. Update the email’s labels to reflect the assigned category.

---

Getting Started

Prerequisites

Step 1: Enable Gmail API

1. Go to the Google Cloud Console and enable the Gmail API.
2. Set up OAuth credentials for a desktop application. Follow these steps to download your google-credentials.json file and place it in your project directory.

Step 2: Install Ollama

Download Ollama to run a local LLM. Once installed, download a model:

ollama pull mistral  

Step 3: Install Dependencies

Install the required Node.js libraries:

ollama pull mistral  

---

Writing the Code

1. Authenticate with Google API

Set up an OAuth connection to Gmail:

npm install -S @google-cloud/local-auth googleapis openai  

2. Create Gmail Labels

Use this function to create labels and retrieve their IDs:

// google-api.js
const fs = require("fs");
const path = require("path");
const { authenticate } = require("@google-cloud/local-auth");
const { google } = require("googleapis");

class GoogleAPI {
  constructor(credentialFilename) {
    this.TOKEN_PATH = path.join(__dirname, `token-${credentialFilename}`);
    this.CREDENTIALS_PATH = path.join(__dirname, credentialFilename);
    this.SCOPES = [
      "https://mail.google.com/",
      "https://www.googleapis.com/auth/gmail.modify",
    ];
  }

  async authorize() {
    const loadSavedCredentials = () => {
      try {
        const content = fs.readFileSync(this.TOKEN_PATH);
        return google.auth.fromJSON(JSON.parse(content));
      } catch {
        return null;
      }
    };

    const saveCredentials = (client) => {
      const keys = JSON.parse(fs.readFileSync(this.CREDENTIALS_PATH));
      fs.writeFileSync(
        this.TOKEN_PATH,
        JSON.stringify({
          type: "authorized_user",
          client_id: keys.installed.client_id,
          client_secret: keys.installed.client_secret,
          refresh_token: client.credentials.refresh_token,
        })
      );
    };

    let client = await loadSavedCredentials();
    if (!client) {
      client = await authenticate({
        scopes: this.SCOPES,
        keyfilePath: this.CREDENTIALS_PATH,
      });
      if (client.credentials) saveCredentials(client);
    }
    return client;
  }
}

module.exports = GoogleAPI;

3. Read Emails

Extract details from message api:

async function createAndGetLabels(labelsToCreate) {
  const google = await getGoogleClient();
  const gmail = google.gmail({ version: "v1" });

  const existingLabels = (await gmail.users.labels.list({ userId: "me" })).data.labels || [];

  const labelsMap = new Map();
  for (const label of labelsToCreate) {
    const existing = existingLabels.find((l) => l.name === label);
    if (existing) {
      labelsMap.set(label, existing.id);
    } else {
      const res = await gmail.users.labels.create({
        userId: "me",
        requestBody: { name: label },
      });
      labelsMap.set(label, res.data.id);
    }
  }
  return labelsMap;
}

4. Decrypt relevants infos

Extract meaningful details from emails:

async function readEmails(gmail, maxResults = 10) {
  const res = await gmail.users.messages.list({ userId: "me", labelIds: ["INBOX"], maxResults });
  return Promise.all(
    res.data.messages.map(async ({ id }) => {
      const email = await gmail.users.messages.get({ userId: "me", id });
      return email.data;
    })
  );
}

5. Use an LLM for Categorization

Integrate Ollama or OpenAI to classify emails:

function extractMailInfos(mail) {
  // Define the headers to extract
  const relevantHeaders = ["Date", "Subject"];

  // Extract and structure the relevant headers
  const headers = mail.payload.headers
    .filter(header => relevantHeaders.includes(header.name))
    .reduce((accumulator, header) => {
      accumulator[header.name] = header.value;
      return accumulator;
    }, {});

  // Add the unique mail ID directly to the headers object
  headers.id = mail.id;

  return headers;
}

---

Putting It All Together

Here’s how everything works together:

async function classifyEmail(prompt) {
  const { OpenAI } = require("openai");
  const openai = new OpenAI({ baseURL: "http://127.0.0.1:11434/v1", apiKey: "not-needed" });

  const response = await openai.chat.completions.create({
    model: "mistral",
    temperature: 0.3,
    messages: [{ role: "user", content: prompt }],
  });

  return response.choices[0].message.content.trim();
}

---

? That’s it! Your inbox is now smarter and more organized.

Go further

But why stop here? Explore more advanced examples for reading email content, sending drafts, and building everything you need for fully automated responses.
For more automation ideas and reusable scripts, check out Znote.

Let’s turn your daily tasks into something fun and efficient! ?

The above is the detailed content of Build an AI-Powered Email Agent with Reusable Code. For more information, please follow other related articles on the PHP Chinese website!