🤖 TG2AI — Telegram-to-AI Data Exporter

TG2AI is an ultra-minimalist, stateless Telegram bot designed to convert any public Telegram channel into clean, AI-ready data formats (Markdown, JSON, CSV) in less than 10 seconds.

Perfect for loading expert context into ChatGPT, Claude Projects, building personal RAG bases, or preparing custom datasets for LLM fine-tuning.

🇷🇺 Бот полностью поддерживает русский язык! Если язык вашего интерфейса Telegram русский, бот автоматически переключится на него.

---

🛑 The Pain: Why not standard Telegram Export?

When you export chat history using the native Telegram Desktop client, you get cluttered HTML or verbose JSON files.

• Bloated Context: Up to 90% of the export consists of service code, system styles, and HTML noise. Feeding this directly to Claude or ChatGPT blows up your context window and burns your token budget.
• No Markdown Structure: Standard exports do not organize posts cleanly for LLM consumption, making note-taking systems like Obsidian difficult to build.

The TG2AI Solution: It scrapes, cleanses, formats, and estimates tokens in 10 seconds. You get clean text, structured code, and active links.

---

🎯 Key Use Cases & Formats

1. 📄 Markdown (.md) — Best for LLM Context & Obsidian

• YAML Frontmatter: Includes channel title, subscribers, timestamp, and a rough token count estimate to help manage your API costs.
• RAG-Ready: Converts Telegram HTML tags (bold, italic, links, code blocks) to standard Markdown. Posts are grouped chronologically by date.

2. 📋 JSON (.json) — Best for Vector DB Ingestion

• Fully structured array of posts containing clean plain text, original markdown structure, views, timestamp, and direct links. Ready to be loaded into Pinecone, ChromaDB, or converted to JSONL for fine-tuning.

3. 📊 CSV (.csv) — Best for Excel & Quantitative Analysis

• Clean table containing fields like views, dates, and text (with secure line breaks escaping). Great for monitoring competitor channels and growth analytics.

---

🚀 Deployment in 3 Steps (Vercel Hobby Free Tier)

This project is designed to run completely stateless and fit comfortably within the Vercel Hobby Free Tier (serverless execution).

1. Clone & Push to GitHub

Create a private or public repository on GitHub, commit your code, and push:

git clone https://github.com/your-username/tg2ai.git
cd tg2ai
npm install

2. Setup Vercel Environment Variables

Import the project into Vercel and add the following Environment Variables:

• TELEGRAM_BOT_TOKEN: Your bot token from @BotFather.
• NEXT_PUBLIC_TELEGRAM_BOT_USERNAME: The username of your bot (e.g., tg2aiibot).
• WEBHOOK_URL: Your Vercel deployment URL (e.g., https://your-project.vercel.app).

3. Set Webhook

Once deployed, open your browser and navigate to: https://your-project.vercel.app/api/set-webhook

You will see: {"ok":true,"result":true,"description":"Webhook was set"}. Your bot is live!

---

🛠️ Local Development & Polling Mode

For fast local testing without setting up tunnels or webhooks:

1. Create a .env file in the root directory:

   TELEGRAM_BOT_TOKEN=your_bot_token
   NEXT_PUBLIC_TELEGRAM_BOT_USERNAME=your_bot_username

2. Start the bot in long-polling mode (it will automatically clear any active webhook for local testing):

   npm run dev      # Runs next dev server
   npx tsx --env-file=.env src/lib/run-polling.ts  # Runs bot listener

---

🤝 Contributing & Wrapping into a "Skill"

We want to make TG2AI the ultimate bridge between Telegram and AI coding agents.

If you are using agentic coding helpers (like Antigravity, Claude Code, or Cline), we are wrapping this utility into a reusable Agentic Skill. This will allow any AI coder to automatically fetch and index Telegram channels directly into a workspace via a simple slash command.

Feel free to open issues, submit Pull Requests, and help us make this tool better!

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.