A Telegram bot that summarises any URL

Goal

Build a Telegram bot that listens for URLs in your DM, fetches the page, sends a 3-line summary back via Claude. Async, uses decorators for message routing, and ships on Railway's free tier.

Why it matters

This is your first real AI product — one API you built, one AI API you called, one deploy. It's also the pattern behind 90% of AI startups you'll ever work at.

Requirements · rubric

Build against this exact list — it is the spec your published project is judged on.

• 01Use `python-telegram-bot` v20+ (async API)
• 02Register a `/start` command and a plain-text URL handler
• 03Detect URLs via regex — reject non-URL messages politely
• 04Fetch page content using `httpx` async client (respect timeouts)
• 05Call Anthropic Messages API to produce a 3-line summary
• 06Return summary + original URL as a formatted reply
• 07Load `TELEGRAM_TOKEN` + `ANTHROPIC_API_KEY` from env — never hardcode
• 08Deploy on Railway (or Fly.io) — provide the public bot handle in your submission
• 09Log errors to stdout; never crash the process on a single bad message

Starter code

Copy this into a new file — or open it in a workspace as bot.py. The TODO parts are yours to fill.

"""
tl;dr bot — summarizes any URL you send.

Env vars:
    TELEGRAM_TOKEN     — from @BotFather
    ANTHROPIC_API_KEY  — from console.anthropic.com

Deploy:
    Set env vars on Railway/Fly.io, then `python bot.py`.
"""

from __future__ import annotations
import os
import re
import sys

# PyRun's workspace runs Python inside your browser (sys.platform is
# "emscripten" there). A Telegram bot needs a server that stays up, a bot
# token, and packages a browser can't install, so in the browser this file
# explains itself instead of crashing. On your machine it runs normally.
IN_BROWSER = sys.platform == "emscripten"

if not IN_BROWSER:
    import httpx
    from anthropic import Anthropic
    from telegram import Update
    from telegram.ext import (
        Application,
        CommandHandler,
        ContextTypes,
        MessageHandler,
        filters,
    )

URL_RE = re.compile(r"https?://[^\s]+")
anthropic = None if IN_BROWSER else Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])


async def start(update: Update, _: ContextTypes.DEFAULT_TYPE) -> None:
    await update.message.reply_text(
        "Send me a URL and I'll send you a 3-line summary."
    )


async def handle_message(update: Update, _: ContextTypes.DEFAULT_TYPE) -> None:
    text = update.message.text or ""
    match = URL_RE.search(text)
    if not match:
        await update.message.reply_text(
            "Send me a URL and I'll summarize it. Nothing else works yet."
        )
        return

    url = match.group(0)
    # TODO: fetch page (respect content-type, timeout, redirects)
    # TODO: extract main text (readability is nice, but text-only is fine)
    # TODO: call anthropic.messages.create with a 3-line summary prompt
    # TODO: reply with formatted summary + source URL
    ...


def main() -> None:
    app = Application.builder().token(os.environ["TELEGRAM_TOKEN"]).build()
    app.add_handler(CommandHandler("start", start))
    app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, handle_message))
    app.run_polling()


if __name__ == "__main__":
    if IN_BROWSER:
        print("This bot runs on a server, not in the browser tab.")
        print("1. pip install python-telegram-bot anthropic httpx")
        print("2. Set TELEGRAM_TOKEN (from @BotFather) and ANTHROPIC_API_KEY")
        print("3. python bot.py  (on your machine, or on Railway / Fly.io)")
        print("You can write and organise the code here; run it there.")
    else:
        main()