Tutorial: Build a Telegram Bot with LLM Replies

This tutorial walks through creating a Telegram bot that replies to every message using an LLM running locally via Ollama. The same pattern applies to Discord, Slack, and WhatsApp with minor config changes.

Prerequisites

• kdeps CLI installed (kdeps version)
• Docker (for running the agent container)
• A Telegram bot token — create one with @BotFather (/newbot)
• Ollama installed locally, or installOllama: true in agentSettings

---

Step 1 — Create the Workflow File

# workflow.yaml
apiVersion: kdeps.io/v1
kind: Workflow

metadata:
  name: telegram-llm-bot
  description: Telegram bot that answers messages with an LLM
  version: "1.0.0"
  targetActionId: reply

settings:
  agentSettings:
    timezone: Etc/UTC
    installOllama: true
    models:
      - llama3.2:3b

  input:
    sources: [bot]
    bot:
      executionType: polling
      telegram:
        botToken: "{{ env('TELEGRAM_BOT_TOKEN') }}"
        pollIntervalSeconds: 1

Key points:

• sources: [bot] enables the bot input subsystem
• executionType: polling keeps the process running and polls Telegram for new messages
• botToken uses the env() expression so the token is never hard-coded
• targetActionId: reply — the workflow ends by executing the reply resource

---

Step 2 — Create the LLM Resource

# resources/llm.yaml
apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: llm
  name: LLM Response

run:
  chat:
    backend: ollama
    model: llama3.2:3b
    messages:
      - role: user
        content: "{{ input('message') }}"

input('message') retrieves the text the Telegram user sent.

---

Step 3 — Create the Reply Resource

The botReply resource sends the text back to the originating platform. In polling mode it calls the platform's API; in stateless mode it writes to stdout. The dispatcher loop continues after this resource returns — no explicit restart is needed.

# resources/reply.yaml
apiVersion: kdeps.io/v1
kind: Resource

metadata:
  actionId: reply
  name: Reply

dependencies:
  - llm

run:
  botReply:
    text: "{{ get('llm') }}"

---

Step 4 — Run the Bot

Export your bot token, then start the workflow:

export TELEGRAM_BOT_TOKEN="1234567890:AAH..."

kdeps run workflow.yaml

You should see:

Bot input sources active:
  • Telegram (polling)

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

Send a message to your bot in Telegram — it replies with the LLM's answer.

---

Step 5 — Add a System Prompt (Optional)

Give your bot a persona by adding a scenario block to the LLM resource:

run:
  chat:
    backend: ollama
    model: llama3.2:3b
    scenario:
      - role: assistant
        prompt: |
          You are Kodi, a helpful AI assistant.
          Keep your answers short and friendly.
          Always respond in the same language as the user.
    messages:
      - role: user
        content: "{{ input('message') }}"

---

Stateless Mode

Stateless mode runs the workflow exactly once from a shell command — no long-running process needed. Useful for cron jobs, CI pipelines, or custom integrations.

Change executionType to stateless:

settings:
  input:
    sources: [bot]
    bot:
      executionType: stateless

Then pipe a JSON message to kdeps run:

echo '{"message":"What is 2+2?","chatId":"42","userId":"u1","platform":"custom"}' \
  | kdeps run workflow.yaml

Output (stdout):

4

Or use environment variables instead of JSON:

export KDEPS_BOT_MESSAGE="What is the capital of France?"
export KDEPS_BOT_PLATFORM="cli"
kdeps run workflow.yaml

---

Adding More Platforms

Extend the bot block to run on Discord and Telegram simultaneously:

settings:
  input:
    sources: [bot]
    bot:
      executionType: polling
      discord:
        botToken: "{{ env('DISCORD_BOT_TOKEN') }}"
      telegram:
        botToken: "{{ env('TELEGRAM_BOT_TOKEN') }}"

The same workflow resources receive messages from both platforms. Use input('platform') to branch if needed:

# resources/reply.yaml
run:
  botReply:
    text: |
      {{ if eq (input('platform')) "discord" }}
      **{{ get('llm') }}**
      {{ else }}
      {{ get('llm') }}
      {{ end }}

---

Full Directory Structure

telegram-llm-bot/
├── workflow.yaml
└── resources/
    ├── llm.yaml
    └── reply.yaml

---

See Also

• Input Sources — All bot platform configs and field reference
• Telegram Bot Example — Ready-to-run example
• Stateless Bot Example — One-shot stdin/stdout example
• LLM Resource — Chat, scenario, backend options