Telegram LLM Bot

A Telegram bot that relays user messages to an LLM and streams responses back using Telegram's sendMessageDraft API for real-time typing animation.

Features

• Long-polling for incoming messages
• LLM response streaming via RubyLLM
• Real-time draft updates with cursor animation (▌) via sendMessageDraft
• Supports OpenAI, Anthropic, and Gemini providers

Requirements

• Ruby 3.4 (managed via mise)
• A Telegram bot token
• An API key for at least one LLM provider

Getting a Telegram Bot Token

1. Open Telegram and search for @BotFather
2. Send /newbot and follow the prompts (choose a name and username)
3. BotFather will reply with your bot token — copy it into TELEGRAM_BOT_TOKEN in your .env

Setup

cp .env.example .env
# Edit .env with your credentials
bundle install

Configuration

Variable	Required	Default	Description
TELEGRAM_BOT_TOKEN	Yes	—	Bot token from @BotFather
OPENAI_API_KEY	One required	—	OpenAI API key
ANTHROPIC_API_KEY	One required	—	Anthropic API key
GEMINI_API_KEY	One required	—	Gemini API key
LLM_MODEL	No	gpt-4o-mini	Model to use (e.g. claude-sonnet-4-5)

Usage

ruby bot.rb

Send any message to your bot and it will stream back an AI-generated response.

How it works

1. Bot polls for updates via Telegram's getUpdates
2. On receiving a message, creates a RubyLLM chat session and calls ask with a streaming block
3. As chunks arrive, accumulated text is sent every ~400ms via sendMessageDraft (same draft_id → Telegram animates the update)
4. Once streaming completes, the full response is sent via sendMessage, replacing the draft