Struct Message 

pub struct Message {
    pub role: Role,
    pub content: Vec<ContentPart>,
}

A single turn in the conversation.

content is always a list of ContentParts. Codecs that accept a string shorthand are responsible for collapsing a single ContentPart::Text to a bare string at encode time.

Message is an open data carrier: codec/runnable internals pattern-match exhaustively against the IR, so the type stays constructable via struct-literal syntax. New IR signals land as additional ContentPart variants (which ContentPart’s #[non_exhaustive] covers) or as new helpers on Message.

Fields

role: Role

Who authored this message.

content: Vec<ContentPart>

One or more content parts. Empty content is permitted (some providers emit empty assistant messages alongside tool calls).

Implementations

impl Message

pub fn new(role: Role, content: Vec<ContentPart>) -> Self

Construct a message with a typed role + content list. Use the role-specific helpers (user / assistant / system / tool_*) for the common single-text-part cases; reach for new when assembling multi-part content (multimodal, tool-use blocks, etc.).

pub fn user(text: impl Into<String>) -> Self

Convenience: user message with a single text part.

pub fn assistant(text: impl Into<String>) -> Self

Convenience: assistant message with a single text part.

pub fn system(text: impl Into<String>) -> Self

Convenience: system message with a single text part.

pub fn tool_result( tool_use_id: impl Into<String>, name: impl Into<String>, output: impl Into<String>, ) -> Self

Convenience: tool message wrapping a tool’s reply to a prior ContentPart::ToolUse. Mirrors LangChain’s ToolMessage(content=…, tool_call_id=…, name=…) shape so the RAG / agent loop reads as a one-line append after each tool call instead of hand-constructing a Message { role: Role::Tool, content: vec![ContentPart::ToolResult { … }] }.

Both tool_use_id and name are required: Anthropic / OpenAI / Bedrock correlate by id, but Gemini’s functionResponse keys by name. Carrying both keeps the IR provider-neutral so a single agent harness works across all four codecs without per-vendor adaptation.

output accepts any string-like — for structured payloads, use Self::tool_result_json and the codec will emit native JSON (or stringify with a LossyEncode warning if the provider lacks structured tool-result support).

pub fn tool_result_json( tool_use_id: impl Into<String>, name: impl Into<String>, output: Value, ) -> Self

Same as Self::tool_result but carries a structured JSON payload. Use when the tool returns objects/arrays the model should reason over without re-parsing a stringified blob.

pub fn tool_error( tool_use_id: impl Into<String>, name: impl Into<String>, output: impl Into<String>, ) -> Self

Same as Self::tool_result but flagged as an error reply. Anthropic and Bedrock surface the is_error flag natively; other codecs prefix the text or emit a LossyEncode warning.

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.

impl<'de> Deserialize<'de> for Message

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for Message

fn eq(&self, other: &Message) -> 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 Message

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 Message

impl StructuralPartialEq for Message