Generate the next message for a prospect using their full state, product context, conversation history, and objection memory.

?mode=generated (default) — backend calls the LLM and returns the message. ?mode=prompt — backend returns the system prompt + user_content + response schema, WITHOUT calling the LLM. The client (OpenClaw plugin / remote MCP) runs its own LLM, then posts the result back with mode=record. This shifts LLM cost off the backend and onto the user's subscription. ?mode=record — client-generated message is logged as an interaction. Requires message in payload. Daily cap still enforced.