Build a Telegram AI Bot in Python for Under ₹1/Day

Telegram bots are the cheapest way to ship an AI project

Telegram's Bot API is free, requires no business verification, and the python-telegram-bot library handles the webhook/polling plumbing for you. Pair it with a budget LLM through AICredits and a hobby-scale bot — a study-group Q&A bot, a personal reminder assistant, a community FAQ bot — costs close to nothing to run.

Setup

pip install python-telegram-bot openai

Get a bot token from @BotFather on Telegram, then:

import logging
from telegram import Update
from telegram.ext import ApplicationBuilder, MessageHandler, ContextTypes, filters
from openai import OpenAI
 
client = OpenAI(base_url="https://api.aicredits.in/v1", api_key="sk-your-aicredits-key")
 
async def handle_message(update: Update, context: ContextTypes.DEFAULT_TYPE):
    user_text = update.message.text
 
    response = client.chat.completions.create(
        model="google/gemini-2.0-flash",
        messages=[
            {"role": "system", "content": "You are a helpful, concise assistant in a Telegram group. Keep replies under 3 sentences."},
            {"role": "user", "content": user_text},
        ],
    )
 
    await update.message.reply_text(response.choices[0].message.content)
 
app = ApplicationBuilder().token("YOUR_TELEGRAM_BOT_TOKEN").build()
app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, handle_message))
app.run_polling()

This runs as a long-running process — deploy it on any small VM, or a free-tier cloud service that supports persistent processes (Telegram's polling model needs the process to stay alive; webhooks are the alternative if you're on a serverless platform).

Adding memory across messages

Telegram bots are stateless by default — each message arrives with no context of prior turns. A simple in-memory conversation history per chat:

conversations: dict[int, list[dict]] = {}
 
async def handle_message(update: Update, context: ContextTypes.DEFAULT_TYPE):
    chat_id = update.message.chat_id
    user_text = update.message.text
 
    history = conversations.setdefault(chat_id, [
        {"role": "system", "content": "You are a helpful, concise assistant."}
    ])
    history.append({"role": "user", "content": user_text})
 
    response = client.chat.completions.create(model="google/gemini-2.0-flash", messages=history)
    reply = response.choices[0].message.content
 
    history.append({"role": "assistant", "content": reply})
    conversations[chat_id] = history[-10:]  # cap history to last 10 messages to bound cost
 
    await update.message.reply_text(reply)

Capping history length matters here specifically — without it, a long-running conversation keeps growing the input token count on every single message, quietly increasing cost per reply over time.

What it actually costs

Assume a small community bot: 200 messages/day, each exchange averaging 400 input tokens (with capped history) and 80 output tokens.

| Model | Daily cost | Monthly cost | |-------|------------|----------------| | Gemini 2.0 Flash | ~₹0.19 | ~₹5.70 | | DeepSeek V3 | ~₹0.20 | ~₹6.00 | | GPT-4o-mini | ~₹0.24 | ~₹7.20 | | Claude 3.5 Haiku | ~₹1.60 | ~₹48 |

For a genuinely hobby-scale bot, this comfortably fits under ₹1/day on any of the budget models — a ₹50 topup lasts months.

Frequently Asked Questions

Should I use polling or webhooks for a Telegram bot?

Polling (run_polling()) is simpler and fine for personal or small-community bots. Webhooks scale better for high-traffic bots but need a public HTTPS endpoint — most hobby deployments don't need this complexity.

How do I keep the bot from responding to every message in a group chat?

Filter on mentions (@your_bot_name) or replies to the bot's own messages, rather than responding to every message in a shared group — check update.message.entities for mention detection.

Can I add voice message support?

Yes — download the voice note via Telegram's file API, then transcribe it with openai/whisper-1 or sarvam/saarika-v2 for Indian languages before passing the text to the chat model.

Get started

• Create a free account and get a key for your bot.
• Related: How to Build a Retry Strategy for LLM API Calls — useful once your bot has real traffic and occasional provider hiccups to handle gracefully.