Bot replies

A bot workflow replies by returning a result: kdeps sends the workflow's result back to the chat that delivered the message - Discord, Slack, Telegram, or stdout in stateless mode. There is no separate reply resource; end the workflow with an apiResponse: (or any resource whose output is text) and set it as targetActionId.

This is workflow mode with bot under settings.input.sources - see Input sources.

message -> workflow (llm -> reply) -> result text -> same chat

Basic usage

# workflow.yaml (abridged)
metadata:
  targetActionId: reply        # its result is the reply
settings:
  input:
    sources: [bot]
    bot:
      executionType: polling
      telegram: {}   # presence enables the platform; credentials are in ~/.kdeps/config.yaml bot_connections

# resources/llm.yaml
actionId: llm
name: Generate Reply
chat:
  model: llama3.2:1b
  prompt: "{{ input('message') }}"

# resources/reply.yaml
actionId: reply
name: Reply
requires: [llm]
apiResponse:
  success: true
  response: "{{ get('llm').message.content }}"   # the text sent to the chat

What is sent

Workflow result	Reply text
apiResponse with a string response	That string
apiResponse with a map or list response	The response as JSON
A chat result (target is the chat resource itself)	message.content
A plain string	The string
Anything else	The result as JSON

An empty result sends nothing. A failed workflow sends nothing and is logged; send a fallback with the resource's onError: if users should get one.

Stateless mode

executionType: stateless reads one message as JSON on stdin and prints the reply on stdout:

echo '{"message":"What is 2+2?","platform":"cli"}' | kdeps run workflow.yaml
# 4

Full example

See the Stateless bot and Telegram bot examples.

See also

• Input sources - configuring settings.input.bot
• API response - the apiResponse: block