Message composer AI (draft / improve)

The AI writing actions in the conversation composer (draft reply, fix spelling, grammar, shorten, expand, and custom instructions) call a single generic AI endpoint with a JSON schema for the reply shape.

Endpoint

POST
/ai-assistant-system

Request (draft reply pattern)

The app builds system_prompt from the current draft and channel context, and sends user_prompt asking for the next outgoing message in JSON.

Request (improve / rewrite pattern)

Same endpoint with source: "ai_conversation_message_composer", a system_prompt that includes the current outgoing draft (and optional channel context), and a user_prompt such as:

  • Fix spelling
  • Fix grammar
  • Improve clarity
  • Shorten
  • Expand
  • Or any custom user-entered instruction

Response (shape used by the app)

json

The client parses ai_response_content as JSON (including handling markdown code fences if present) into html_content, text_content, explanation, and suggestions.

Body fields

FieldTypeDescription
system_promptStringFull system instructions (includes transcript summary and draft where applicable).
user_promptStringTask for the model (draft, fix, custom).
sourceStringAlways ai_conversation_message_composer for this feature.
organization_idNumberOrganization scope.
ai_schemaObjectJSON-schema-like description for structured model output.

Notes

  • The composer loads recent inbound messages from List conversation messages (limit 50, excluding reactions) to build the transcript section of system_prompt.
  • This endpoint is not tied to a specific saved AI assistant record; it is a generic completion used only for drafting in the composer.

Report an issue with this documentation

Please log in to report issues with our documentation.