Struct TurnResult 

pub struct TurnResult {
    pub reply: String,
    pub message_id: String,
    pub invoked_knowledge_search: bool,
    pub citations: Vec<Citation>,
    pub usage: Option<TurnUsage>,
    pub next_step_id: Option<String>,
}

The terminal outcome of a streamed turn.

Fields

reply: String

The agent’s final natural-language reply.

message_id: String

The id of the persisted outbound (agent) message, for eventual_response.

invoked_knowledge_search: bool

True if any knowledge_search tool call ran this turn (diagnostics).

citations: Vec<Citation>

The sources that grounded this turn (the auto-injected context + every knowledge_search result), deduped by id and capped. Carried onto the eventual_response’s citations. Empty when nothing was retrieved.

usage: Option<TurnUsage>

The turn’s token-accounting + cost, captured from the engine’s terminal AgentEvent::Completed. Carried onto the eventual_response’s usage object so clients accumulate live session cost. None when the engine reported no Completed event (e.g. an offline mock turn).

next_step_id: Option<String>

The conversation-workflow step id after this turn’s judge ran. Some only when the turn had a WorkflowTurn; the caller persists it onto the session so the next turn resumes on the right step. Equals the incoming step when the judge did not advance (criteria not met, terminal step, or a judge failure — never freezes, never crashes the turn).