Skip to main content

AI Response

AI is invoked and generates a response when all of the following conditions are met:

  1. “Auto response by AI” is enabled in settings.
  2. A text message is received from the user.
  3. No message response is generated by any triggers.

Workflow​

Available LLMs​

LLMs are called via APIs. The OpenAI API or any API compatible with OpenAI API can be used.
Requests are sent to the configured OpenAI API Endpoint using the selected API Key and Model, so any chat/completions-compatible endpoint can be used.

Function Calling​

In addition to the default functions, you can add custom functions via filter hooks.
For instructions on adding functions, see the Filter hooks page. Even when Function Calling is enabled, only the functions selected in settings are sent to the model.
If a function definition has a role, only users with the required capability can call it.
Some functions return response_mode: direct, in which case the function's message is used directly instead of going through another AI generation step.

System Prompt​

You can pre-specify the AI's purpose, role, response style, and other behaviors.
You can embed the current time, user information, and incoming event data using the Twig template engine notation.
webhook contains the received event, including postback data and params.
Example:)

Current time: {{ "now"|date("Y-m-d H:i:s") }}
WP user ID: {{ user.data.ID }}
User name: {{ user.data.display_name }}
Email address: {{ user.data.user_email }}
postback data: {{ webhook.postback.data }}
postback message id: {{ webhook.postback.params.slc_message_id }}

Context​

  • Previous conversation logs are sent to the model up to the configured number of context items.
  • If the user replies to a quoted message, the quoted message is also added to the prompt context.
  • Image messages are passed to the model as image_url.
  • Audio messages in history are handled as URL strings in context, not as native audio input.

Other Settings​

  • Temperature controls response diversity. Higher values generally produce more varied output.
  • Max tokens sets the maximum tokens used for a single response. -1 uses the model limit.
  • Daily usage limits can be configured separately for linked and unlinked users.
  • When the limit is exceeded, the configured limit message is returned.

Multimodal Input​

  • Supports image messages.
  • To ask about an image, reply to that image message and send your text question.
  • Sending plain text without replying to the image will not use the image as context.