Enum MultiTurnStreamItem 

#[non_exhaustive]
pub enum MultiTurnStreamItem<R> {
    StreamAssistantItem(StreamedAssistantContent<R>),
    ToolExecutionCommitted {
        tool_call: ToolCall,
        internal_call_id: String,
    },
    StreamUserItem(StreamedUserContent),
    CompletionCall(CompletionCall),
    ModelTurnRetried {
        turn: usize,
    },
    FinalResponse(PromptResponse),
}

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

StreamAssistantItem(StreamedAssistantContent<R>)

A streamed assistant content item — the content the model emitted: text/reasoning deltas, tool-call deltas, and, when the model turn is committed, the complete StreamedAssistantContent::ToolCall for each tool call Rig routes to execution. Such a call is reported here whether or not the tool body ultimately runs (a hook skip still reports it); it is not an execution-lifecycle event (see ToolExecutionCommitted).

Two kinds of model tool call are not re-emitted as a complete ToolCall item here (their arguments still stream as tool-call deltas): a call rejected and handled by invalid-tool-call recovery (surfaced via that recovery path), and a structured-output Tool-mode output-tool call, which finalizes the run directly — its structured result is surfaced in the FinalResponse rather than as a completed ToolCall item.

ToolExecutionCommitted

Confirmation that Rig executed and committed a tool call. This is not a real-time start notification: it is surfaced together with its ToolResult only after the whole batch settles successfully. Use tool hooks for live host-side start/result observation.

This item is emitted only for a tool whose body actually ran (it passed its ToolCall hook checks), never for a call dropped by a sibling’s termination, skipped by a hook, or resolved by invalid-call recovery. Correlate it with the model call and result through internal_call_id.

Fields

tool_call: ToolCall

The tool call as executed: the model’s call with any ToolCallAction::Rewrite hook rewrite applied (so a redaction rewrite is reflected here, not leaked). The model’s original call is reported via StreamAssistantItem.

internal_call_id: String

Rig-generated id correlating this execution with the model tool call (StreamedAssistantContent::ToolCall::internal_call_id) and the resulting StreamedUserContent::ToolResult.

StreamUserItem(StreamedUserContent)

A streamed user content item: the result of an executed (or hook-skipped) tool call. The tool batch commits and surfaces atomically at every tool_concurrency (including the sequential default): results are surfaced (in call order) only after the whole batch settles successfully — a run that terminates mid-batch surfaces no successful tool results.

CompletionCall(CompletionCall)

Details for one successfully completed completion request made by this agent stream.

This is emitted when a provider call finishes. Usage is the provider’s final usage for that completion request when available; it is not incremental per streamed token.

match item {
    MultiTurnStreamItem::CompletionCall(completion_call) => {
        // Zero-valued usage means the provider reported no metrics.
        if completion_call.usage.has_values() {
            let context_tokens = completion_call.usage.input_tokens;
        }
    }
    _ => {}
}

ModelTurnRetried

The completed model turn was rejected by a hook for retry.

Text and reasoning deltas emitted for this turn were provisional. A consumer should discard or visually reset output associated with turn. A subsequent attempt is made only if the run’s total model-call budget permits it.

Fields

turn: usize

One-based model-call index of the rejected turn.

FinalResponse(PromptResponse)

The final result from the stream: the unified PromptResponse shared with the blocking surface.

Implementations

impl<R> MultiTurnStreamItem<R>

pub fn final_response( content: OneOrMany<AssistantContent>, aggregated_usage: Usage, ) -> Self

pub fn final_response_with_history( content: OneOrMany<AssistantContent>, aggregated_usage: Usage, history: Option<Vec<Message>>, ) -> Self

Trait Implementations

impl<R: Clone> Clone for MultiTurnStreamItem<R>

fn clone(&self) -> MultiTurnStreamItem<R>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<R: Debug> Debug for MultiTurnStreamItem<R>

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

Formats the value using the given formatter.

impl<'de, R> Deserialize<'de> for MultiTurnStreamItem<R>

where R: Deserialize<'de>,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl<R> Serialize for MultiTurnStreamItem<R>

where R: Serialize,

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.