Get Conversation

Cost + token usage for a conversation (or a single sub-execution).

Sourced from the same monitoring rollup the api service exposes at /v1/tasks/{task_id}/llm_usage. Flow:

chat-backend → agent-controller (in-process call) → deployment-manager (agents_activity_proxy) → agents service (/{agent_id}/activity/threads/llm_usage:aggregate) → MongoDB monitoring docs + LLM tier table

Returned with the conversation as a convenience so the UI doesn't have to issue a second roundtrip per render.

Triple-shape entry for every sub-execution spawned by this conversation.

Per-message feedback rows recorded against the parent execution.