Enum Message 

pub enum Message {
    User(Vec<UserContent>),
    Assistant {
        content: Option<String>,
        reasoning: Option<String>,
        tool_calls: Vec<ToolCall>,
        provider_data: Option<Value>,
    },
    ToolResult {
        call_id: String,
        content: Vec<Content>,
    },
}

A single turn in a conversation. Every variant carries exactly the fields it needs — no invalid states are representable.

Variants

User(Vec<UserContent>)

A message from the human side, supporting text and images.

Assistant

A message produced by the model. content and tool_calls may both be present; content may be absent when the model only emits tool calls.

Fields

content: Option<String>

reasoning: Option<String>

Provider-specific chain-of-thought / reasoning text, if any.

tool_calls: Vec<ToolCall>

provider_data: Option<Value>

Opaque per-turn state emitted by the provider (e.g. Anthropic thinking blocks with signatures). Populated from LlmEvent::AssistantState and round-tripped verbatim by the same provider’s request serializer. Always None for providers that don’t emit state.

ToolResult

The result of a tool invocation, keyed by the call’s ID.

Fields

call_id: String

content: Vec<Content>

Implementations

impl Message

pub fn estimate_tokens(&self) -> usize

Estimate the number of tokens in this message using tiktoken.

Note: This is an estimation. Different providers have slightly different tokenization rules and overheads for message metadata (role, name, etc.).

Trait Implementations

impl Clone for Message

fn clone(&self) -> Message

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Message

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

Formats the value using the given formatter.