Stateless Bot Example

A one-shot bot workflow that reads a message from stdin (or environment variables), processes it with an LLM, and writes the reply to stdout. No long-running process — just pipe in a message and get a response.

Features

• Single execution: start, process one message, exit
• Input via stdin JSON or environment variables
• Output via stdout — easy to compose with shell scripts, cron, CI

Usage

JSON stdin

echo '{"message":"What is the capital of France?","chatId":"42","userId":"u1","platform":"cli"}' \
  | kdeps run workflow.yaml

Output:

Paris is the capital of France.

Environment Variables

export KDEPS_BOT_MESSAGE="Summarize the theory of relativity in one sentence."
export KDEPS_BOT_PLATFORM="cli"
kdeps run workflow.yaml

From a Shell Script

#!/bin/bash
REPLY=$(echo '{"message":"Hello!"}' | kdeps run /path/to/stateless-bot/workflow.yaml)
echo "Bot replied: $REPLY"

Cron Job (daily greeting)

0 9 * * * KDEPS_BOT_MESSAGE="Good morning! What's a fun fact for today?" \
          kdeps run /path/to/stateless-bot/workflow.yaml >> /var/log/bot-greetings.log

Input Format

Method	Format
stdin JSON	{"message":"...","chatId":"...","userId":"...","platform":"..."}
Env vars	KDEPS_BOT_MESSAGE, KDEPS_BOT_CHAT_ID, KDEPS_BOT_USER_ID, KDEPS_BOT_PLATFORM

Only message (or KDEPS_BOT_MESSAGE) is required. All other fields are optional.

Structure

stateless-bot/
├── workflow.yaml          # Stateless bot source config
└── resources/
    ├── llm.yaml           # LLM chat — processes input('message')
    └── reply.yaml         # Writes LLM reply to stdout

See Also

• Telegram Bot Example — Long-running polling bot
• Input Sources Documentation
• Bot Tutorial