Enum LlmResponse 

pub enum LlmResponse {
    Start {
        message_id: String,
    },
    Text {
        chunk: String,
    },
    Reasoning {
        chunk: String,
    },
    EncryptedReasoning {
        id: String,
        content: String,
    },
    ToolRequestStart {
        id: String,
        name: String,
    },
    ToolRequestArg {
        id: String,
        chunk: String,
    },
    ToolRequestComplete {
        tool_call: ToolCallRequest,
    },
    Done {
        stop_reason: Option<StopReason>,
    },
    Error {
        message: String,
    },
    Usage {
        tokens: TokenUsage,
    },
}

A streaming response event from an LLM provider.

Providers return an LlmResponseStream that yields a sequence of these events as the model generates its response. The stream follows a defined lifecycle:

Start -> (Text | Reasoning | EncryptedReasoning | ToolRequest*)* -> Usage -> Done

Text generation

• Start – Stream opened, contains the message_id.
• Text – A chunk of generated text.
• Reasoning – A chunk of the model’s chain-of-thought reasoning (visible summary).
• EncryptedReasoning – Opaque encrypted reasoning content (model-specific, can be replayed to the same model in future turns).

Tool calling

Tool calls are streamed in three phases:

1. ToolRequestStart – The model begins a tool call (id + name).
2. ToolRequestArg – Argument JSON arrives in chunks (same id).
3. ToolRequestComplete – The fully assembled ToolCallRequest is ready to execute.

Multiple tool calls can be interleaved in a single response.

Completion

• Usage – Token usage statistics, carried as a TokenUsage. Required: input_tokens, output_tokens. Optional sub-categories (each filled in by the providers that expose it):
  • cache_read_tokens – prompt tokens served from cache (Anthropic, Bedrock, OpenAI Chat/Responses/Codex, OpenAI-compat).
  • cache_creation_tokens – prompt tokens written to cache on this turn (Anthropic cache_creation_input_tokens, Bedrock cache_write_input_tokens, OpenRouter cache_write_tokens).
  • input_audio_tokens – input audio tokens, subset of input_tokens (OpenAI Chat, OpenRouter).
  • input_video_tokens – input video tokens, subset of input_tokens (OpenRouter).
  • reasoning_tokens – reasoning/thinking tokens, subset of output_tokens (OpenAI Chat/Responses/Codex, OpenRouter).
  • output_audio_tokens – output audio tokens, subset of output_tokens (OpenAI Chat, OpenRouter).
  • accepted_prediction_tokens – subset of output_tokens for predicted-output models (OpenAI Chat, OpenRouter).
  • rejected_prediction_tokens – subset of output_tokens for predicted-output models (OpenAI Chat, OpenRouter).
• Done – Stream complete, with an optional StopReason.
• Error – An error occurred during generation.

Convenience constructors

Each variant has a corresponding constructor method (e.g. LlmResponse::text("chunk"), LlmResponse::done()) to simplify test fixture construction.

Variants

Start

Fields

message_id: String

Text

Fields

chunk: String

Reasoning

Fields

chunk: String

EncryptedReasoning

Fields

id: String

content: String

ToolRequestStart

Fields

id: String

name: String

ToolRequestArg

Fields

id: String

chunk: String

ToolRequestComplete

Fields

tool_call: ToolCallRequest

Done

Fields

stop_reason: Option<StopReason>

Error

Fields

message: String

Usage

Fields

tokens: TokenUsage

Implementations

impl LlmResponse

pub fn start(message_id: &str) -> Self

pub fn text(chunk: &str) -> Self

pub fn reasoning(chunk: &str) -> Self

pub fn encrypted_reasoning(id: &str, encrypted: &str) -> Self

pub fn tool_request_start(id: &str, name: &str) -> Self

pub fn tool_request_arg(id: &str, chunk: &str) -> Self

pub fn tool_request_complete(id: &str, name: &str, arguments: &str) -> Self

pub fn usage(input_tokens: u32, output_tokens: u32) -> Self

pub fn done() -> Self

pub fn done_with_stop_reason(stop_reason: StopReason) -> Self

Trait Implementations

impl Clone for LlmResponse

fn clone(&self) -> LlmResponse

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LlmResponse

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for LlmResponse

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for LlmResponse

fn eq(&self, other: &LlmResponse) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for LlmResponse

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Eq for LlmResponse

impl StructuralPartialEq for LlmResponse