Choose who answers: AI, Co-Pilot or Human

Decides who writes the next reply on this one conversation.

• ai — the AI answers by itself, no operator involved.
• copilot — the AI stops answering and instead writes drafts the operator reviews and sends (POST /conversations/{uuid}/drafts).
• human — the AI stops answering and writes nothing; the operator types the reply.

When to use it. When an operator takes a conversation off the AI — a complaint, a negotiation, anything the AI should not be handling alone — and again when they hand it back. Sending ai returns the thread to automatic answering.

Why it is stored server-side. The next operator to open the conversation sees the same mode, on any device: the state belongs to the conversation, not to one browser tab.

Relationship to take_over. copilot and human both set take_over to true — the flag every channel checks before letting the AI reply — and ai clears it. Both fields are returned on the conversation object.

Starting value. A new conversation follows the mode of its queue’s channel assignment (autopilot / copilot / human_first). This endpoint is the per-conversation override; it does not change the queue.

Who can call it. A workspace API token, or a per-user JWT holding loop.conversations.mode. JWTs only reach conversations in their active queues (404 otherwise). A JWT is attributed to its own user; a workspace token carries no identity, so pass user_email to record who switched the mode.