Telegram Bot Example

A Telegram bot that receives user messages and replies using a local LLM (Ollama llama3.2:3b).

Features

• Polls Telegram for new messages every second
• Sends each message to a local LLM with a custom persona ("Kodi")
• Replies to the user automatically on Telegram

Prerequisites

• A Telegram bot token — create one with @BotFather (/newbot)
• Docker or a local kdeps runtime

Setup

1. Get a Bot Token

Talk to @BotFather on Telegram:

/newbot
→ Name: My Kodi Bot
→ Username: my_kodi_bot
→ Token: 1234567890:AAH...

2. Run the Bot

export TELEGRAM_BOT_TOKEN="1234567890:AAH..."

# From the examples/telegram-bot directory
kdeps run workflow.yaml

# Or from the project root
kdeps run examples/telegram-bot/workflow.yaml

You should see:

Bot input sources active:
  • Telegram (polling)

Starting bot runners... (press Ctrl+C to stop)

3. Chat with the Bot

Open Telegram, find your bot by its username, and send it a message. It will reply using the LLM.

Structure

telegram-bot/
├── workflow.yaml              # Bot source config, Telegram credentials, Ollama model
├── components/
│   └── botreply/
│       └── component.yaml     # .komponent: sends reply through the bot session
└── resources/
    └── llm.yaml               # LLM chat resource — receives input('message'), applies persona

The botreply component encapsulates the run.botReply executor and is auto-loaded from the components/ directory when the workflow is parsed. This makes the component independently distributable — install it once with kdeps component install botreply and reuse it across multiple Telegram bot workflows.

To package the component as a shareable .komponent archive:

kdeps package components/botreply --output components/

Key Expressions

Expression	Description
input('message')	The text the Telegram user sent
input('chatId')	Telegram chat ID (for targeted replies)
input('userId')	Telegram user ID
input('platform')	Always "telegram" for this bot
get('llm')	The LLM-generated reply text

Customization

Use a Different Model

# workflow.yaml
agentSettings:
  models:
    - llama3.1:8b     # Smarter, slower

# resources/llm.yaml
chat:
  model: llama3.1:8b

Change the Persona

Edit the scenario block in resources/llm.yaml:

scenario:
  - role: assistant
    prompt: |
      You are Aria, a customer support agent for Acme Corp.
      Help users with order status, returns, and product questions.

Add Discord Support

Add a discord block to workflow.yaml:

bot:
  executionType: polling
  discord:
    botToken: "{{ env('DISCORD_BOT_TOKEN') }}"
  telegram:
    botToken: "{{ env('TELEGRAM_BOT_TOKEN') }}"

See Also

• Stateless Bot Example — One-shot stdin/stdout execution
• Input Sources Documentation
• Bot Tutorial
• Voice Assistant Example — Audio/microphone input with TTS output