Struct AgentStep 

pub struct AgentStep {
    pub message: Message,
    pub usage: Usage,
    pub finish_reason: Option<FinishReason>,
    pub tool_calls: Vec<ToolCall>,
    pub tool_results: Vec<HistoryEntry>,
}

Data record of one LLM round (one model call + tool dispatch).

Carries only what downstream consumers actually read: the assistant message, token usage, the finish reason, and the tool calls / results. No synthesized wire response — the old AgentStep.response: Response field was a parallel type that only served to hold usage and the final text content. Those two fields are now on the step directly.

Fields

message: Message

The assistant message produced by this step.

usage: Usage

Token usage reported by the provider (zero if not reported).

finish_reason: Option<FinishReason>

Why the model stopped generating (if reported).

tool_calls: Vec<ToolCall>

Tool calls made in this step (if any).

tool_results: Vec<HistoryEntry>

Results from tool executions as history entries.

Trait Implementations

impl Clone for AgentStep

fn clone(&self) -> AgentStep

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AgentStep

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.