Struct ChatRequest 

pub struct ChatRequest {
    pub system: String,
    pub messages: Vec<Message>,
    pub tools: Option<Vec<Tool>>,
    pub max_tokens: u32,
    pub max_tokens_explicit: bool,
    pub session_id: Option<String>,
    pub cached_content: Option<String>,
    pub thinking: Option<ThinkingConfig>,
    pub tool_choice: Option<ToolChoice>,
    pub response_format: Option<ResponseFormat>,
}

Fields

system: String

messages: Vec<Message>

tools: Option<Vec<Tool>>

max_tokens: u32

max_tokens_explicit: bool

Whether max_tokens was explicitly configured by the caller.

session_id: Option<String>

Optional session identifier for provider-side prompt caching or routing.

cached_content: Option<String>

Optional provider-managed cached content reference.

This currently maps to Gemini / Vertex AI cachedContent handles.

thinking: Option<ThinkingConfig>

Optional extended thinking configuration.

tool_choice: Option<ToolChoice>

Optional constraint on tool usage.

When None the provider’s default behaviour applies (typically auto).

response_format: Option<ResponseFormat>

Optional request for the final answer to be constrained to a JSON Schema.

When Some, the provider maps this to its native JSON-mode / structured-output capability (or a tool-forcing fallback) and the runtime validates the final output against the schema. When None (default) the model responds freely.

Implementations

impl ChatRequest

pub const DEFAULT_MAX_TOKENS: u32 = 4096

Default token budget used by ChatRequest::new when the caller does not set one explicitly. Providers clamp this to their own ceiling.

pub fn new(system: impl Into<String>, messages: Vec<Message>) -> ChatRequest

Build a request from a system prompt and a message list, leaving every optional knob at its default.

This is the ergonomic counterpart to the (still-public) struct literal: the common case only needs system + messages, so callers no longer have to spell out the eight None/default fields. Layer optional settings on with the chainable with_* setters:

use agent_sdk_foundation::llm::{ChatRequest, Message, ToolChoice};

let req = ChatRequest::new("You are helpful.", vec![Message::user("Hi")])
    .with_max_tokens(1024)
    .with_tool_choice(ToolChoice::Auto);

pub fn with_tools(self, tools: Vec<Tool>) -> ChatRequest

Set the tool list the model may call.

pub const fn with_max_tokens(self, max_tokens: u32) -> ChatRequest

Set the maximum output-token budget (marks it as explicitly configured).

pub fn with_session_id(self, session_id: impl Into<String>) -> ChatRequest

Set the session identifier (provider-side prompt caching / routing).

pub const fn with_thinking(self, thinking: ThinkingConfig) -> ChatRequest

Set the extended-thinking configuration.

pub fn with_tool_choice(self, tool_choice: ToolChoice) -> ChatRequest

Constrain tool usage (defaults to the provider’s auto when unset).

pub fn with_response_format( self, response_format: ResponseFormat, ) -> ChatRequest

Request the final answer be constrained to the given JSON-Schema ResponseFormat (structured output).

Trait Implementations

impl Clone for ChatRequest

fn clone(&self) -> ChatRequest

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ChatRequest

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

Formats the value using the given formatter.