How to Build a Telegram AI Bot That Summarizes Links for Free (Full Code)

The Problem Nobody Warns You About

Group chats are where information goes to die. Someone drops a 3,000-word article, everyone says "saving this for later", and nobody ever reads it.

I got tired of it, so I built a Telegram bot that does one thing well: you send it a link, it replies with a clean summary. No webhooks from third-party services, no paid API keys, no server. Here's the full code and the exact steps, all free.

What You Need

• A Telegram account and the @BotFather bot to create your bot token (2 minutes)
• Python 3.9+
• A free Groq API key (they have a generous free tier for LLM calls: I used their small model for this because summaries don't need a flagship)
• python-telegram-bot and httpx (or requests)

Step 1: Create the Bot

Open @BotFather in Telegram, send /newbot, pick a name and username. You get a token that looks like:

7123456789:AAHf0xJ9kLmNopQrStUvWxYz

Keep it secret. It's the key to your bot.

Step 2: The Summarizer

The core is simple: fetch the link, strip the HTML, truncate, and ask the LLM to summarize. I use Groq's free tier because it's fast and the free limit is generous.

import httpx

GROQ_KEY = "your-groq-key"
MODEL = "llama-3.1-8b-instant"  # small + fast, plenty for summaries

def summarize(url: str) -> str:
    # Fetch the page
    html = httpx.get(url, follow_redirects=True, timeout=30).text

    # Crude but effective HTML stripping
    import re
    text = re.sub(r"<script.*?</script>|<style.*?</style>", "", html, flags=re.S)
    text = re.sub(r"<[^>]+>", " ", text)
    text = re.sub(r"\s+", " ", text).strip()[:4000]

    resp = httpx.post(
        "https://api.groq.com/openai/v1/chat/completions",
        headers={"Authorization": f"Bearer {GROQ_KEY}"},
        json={
            "model": MODEL,
            "messages": [
                {"role": "system", "content": "Summarize the article in 4 bullet points. Plain language, no filler."},
                {"role": "user", "content": text},
            ],
            "temperature": 0.3,
        },
        timeout=60,
    )
    return resp.json()["choices"][0]["message"]["content"]

That's the whole intelligence layer. Four bullets, plain language, done.

Step 3: Wire It Into Telegram

from telegram import Update
from telegram.ext import Application, CommandHandler, MessageHandler, filters

TOKEN = "your-bot-token"

async def start(update: Update, _):
    await update.message.reply_text("Send me any link and I'll summarize it.")

async def handle(update: Update, _):
    url = update.message.text.strip()
    if not url.startswith(("http://", "https://")):
        await update.message.reply_text("That doesn't look like a link. Send me a URL.")
        return
    await update.message.reply_text("Summarizing...")
    try:
        await update.message.reply_text(await summarize(url))
    except Exception as e:
        await update.message.reply_text(f"Failed: {e}")

app = Application.builder().token(TOKEN).build()
app.add_handler(CommandHandler("start", start))
app.add_handler(MessageHandler(filters.TEXT, handle))
print("Bot running...")
app.run_polling()

Step 4: Run It (Two Options)

Option A: your laptop. python bot.py and it works while the terminal is open.

Option B: free forever. Deploy to a free tier that supports long-running processes (a free cloud VM or a free PaaS with a sleep-free plan). Polling means no webhook config needed, just a process that stays alive.

The Gotchas I Hit

1. HTML stripping is never perfect. Some pages are JS-rendered and return almost nothing. For those, I fall back to the page's meta description, which is better than nothing.
2. Rate limits are real. Groq's free tier allows a few thousand requests per day, which is fine for a group bot. If you exceed it, add a tiny delay or cache repeated links in a dict.
3. Don't let the bot reply in huge threads. Summaries should be short. If a page is enormous, the 4,000-char truncation keeps responses snappy.
4. Never log the full URL contents. Some links contain private data. Summarize, don't store.

Make It Yours

Once the basics work, the upgrades are obvious: summarize YouTube transcripts, accept article text pasted directly, or have it tag the summary with a category. Each one is a few lines on top of this core.

The pattern here is the same one I use across my whole automation stack: fetch, transform, deliver. Boring, reliable, free. If you want more patterns like this, the free guides on my site cover news bots, automation stacks, and prompt systems that run for $0.

This bot is a 30-line version of the kind of pipeline I document in full on the free Apex Nexus learning hub. Built with the same philosophy: cheap models, simple code, no unnecessary moving parts.