Example: Telegram LLM bot

This workflow runs in workflow mode (kdeps run) as a Telegram bot - it polls for messages and replies with a local LLM response (llamafile, no server install). Two resources: llm calls the model, reply returns the answer, which kdeps sends back to the chat.

Files

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

workflow.yaml

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

name: telegram-llm-bot
description: Telegram bot that answers messages with an LLM
version: "1.0.0"
targetActionId: reply
settings:
  agentSettings:
    timezone: Etc/UTC

  input:
    sources: [bot]
    bot:
      executionType: polling
      telegram:
        pollIntervalSeconds: 1  # credentials go in ~/.kdeps/config.yaml bot_connections.telegram

resources/llm.yaml

actionId: llm
name: LLM Response
chat:
  messages:
    - role: user
      content: "{{ input('message') }}"

resources/reply.yaml

actionId: reply
name: Reply
requires:
  - llm

apiResponse:                 # in bot mode the workflow result is the reply
  success: true
  response: "{{ get('llm').message.content }}"   # sent as the reply text

Running

export TELEGRAM_BOT_TOKEN="1234567890:AAH..."
kdeps run workflow.yaml

See also

• Input sources - Full platform config reference