I Built an AI Assistant That Reads My Emails and Replies. All Running in the Browser

Last week I asked my AI assistant to check my emails. It opened Gmail, read through my inbox, summarized what was waiting for me, and when I pointed at one email and said "reply to this one," it drafted and sent a response. All of this happened in my browser. No backend server touched my data. No API key left my machine.

This is izan.io (https://izan.io) — and it's fully open source.

The Problem With Every AI Chat App

You've probably used ChatGPT, Claude, or one of the dozen AI chat interfaces out there. They all share the same limitations:

Locked to one provider. Want to try Gemini after using GPT-4? New app, new subscription, new interface.

Your data lives on their servers. Every conversation, every API key, stored somewhere you don't control.

They can't do anything. They talk. That's it. Ask them to check your email or book a flight, and you get a polite refusal.

izan.io solves all three.

What is izan.io?

izan.io is an open-source AI agent platform that runs entirely in your browser. Here's what makes it different:

17+ LLM Providers, One Interface

OpenAI, Google Gemini, Claude, Mistral, Groq, DeepSeek, Llama via Ollama, and more — all accessible from the same chat interface. Bring your own API key, switch models mid-conversation, compare outputs. No subscriptions, no middleman.

Your Data Never Leaves Your Browser

API keys are stored in IndexedDB. Conversations are stored in IndexedDB. There is literally no backend database. When you chat with GPT-4, your browser talks directly to OpenAI's API. izan.io never sees your messages.

Agents That Actually Do Things

This is the big one. izan.io agents don't just generate text — they use tools. And with browser macros, those tools can be anything you can do in a browser.

The Gmail Demo: AI Meets Browser Automation

Let me walk you through what happened in the demo:

Step 1: I had Gmail open in my browser session, with izan.io's Chrome extension active.

Step 2: I told the agent: "Check my emails."

Step 3: The agent used browser automation to navigate Gmail, read through my inbox, and came back with a summary of my recent emails — senders, subjects, and snippets.

Step 4: I picked one email and said: "Reply to this one." The agent composed a contextual reply and sent it.

Step 5: I verified the sent reply — it was there in my sent folder, exactly as the agent composed it.

No Gmail API setup. No OAuth flow. No server-side integration. The agent literally used the browser like a human would — clicking, reading, typing — but faster.

How Is This Possible?

izan.io uses the Model Context Protocol (MCP) combined with a Chrome extension that can record and replay browser interactions:

1. Record a macro: Use the visual recorder to capture browser actions — clicks, typing, scrolling, navigation. No code needed.
2. Macro becomes a tool: The recorded actions are automatically converted into an MCP tool definition.
3. Agent uses the tool: When you ask the agent to do something, it calls the macro tool just like any other function call. The browser automation replays your recorded steps.

Think of it as giving your AI agent hands to use any website.

Multi-Agent Orchestration

Single agents are powerful. Multiple agents working together are something else.

izan.io supports multi-agent orchestration — agents can call other agents as tools, up to 3 levels deep. A general assistant can delegate domain-specific questions to specialized agents, each with their own tools and system prompts.

For example:

• Research Agent gathers information from multiple sources
• Writing Agent takes the research and drafts content
• Review Agent checks the output for quality

All triggered from a single user message.

80+ Built-in Agents

izan.io comes with 80+ pre-configured agents covering:

• Productivity: Gmail, Google Sheets, Notion, Jira, Trello, GitHub
• Social Media: Twitter/X, Instagram, LinkedIn, Reddit, Discord
• Research: Google Scholar, ArXiv, Wikipedia, HackerNews
• E-commerce: Amazon, eBay, Shopify, Booking.com
• Finance: Yahoo Finance, CoinMarketCap
• And many more...

Each agent has curated system prompts and tool configurations. Use them as-is or customize them to fit your workflow.

The Tech Stack (For the Curious)

• Frontend: React 19 + React Router 7 + Vite + Tailwind CSS 4
• State: Zustand stores for chat, agents, MCP, and models
• Database: IndexedDB via Dexie.js — everything client-side
• MCP Servers: Run in-browser using postMessage-based transport
• Chrome Extension: Records macros, replays via Chrome DevTools Protocol
• Infrastructure: AWS CDK (S3 + CloudFront) — static hosting only
• Monorepo: npm workspaces + Turborepo

The architecture is intentionally simple. No backend server processing your messages. No database storing your conversations. The Lambda function that exists is purely a CORS proxy for connecting to custom MCP servers — it doesn't see or store any content.

Getting Started

1. Go to izan.io (https://izan.io)
2. Paste an API key from any supported provider
3. Start chatting

That's it. No account creation, no email verification, no credit card.

For browser automation, install the Chrome extension and record your first macro in under a minute.

It's Open Source

izan.io is licensed under AGPL-3.0. The entire codebase — frontend, agents, MCP servers, Chrome extension, infrastructure — is open.

GitHub: https://github.com/ekingunoncu/izan.io

If you find this useful, a star on GitHub would mean a lot. It helps others discover the project and motivates continued development.

izan.io is built by a small team that believes AI tools should be open, private, and extensible. If you have ideas, feedback, or want to contribute, open an issue or PR — we'd love to hear from you.