Struct CompletionRequest 

#[non_exhaustive]
pub struct CompletionRequest {
    pub model: String,
    pub system: Option<String>,
    pub messages: Vec<Message>,
    pub tools: Vec<ToolSpec>,
    pub tool_choice: ToolChoice,
    pub response_format: Option<JsonSchema>,
    pub max_tokens: Option<u32>,
    pub temperature: Option<f32>,
    pub stop: Vec<String>,
    pub web_search: bool,
}

Top-level request to an LLM provider.

Construct via CompletionRequest::new, then populate fields directly.

#[non_exhaustive]: provider-shaped sampling fields (top_p, seed, …) will be added over time; build through new so such additions stay non-breaking.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

model: String

The model identifier (e.g. "fast-2", "reasoning-pro").

system: Option<String>

Optional system prompt text prepended before the conversation.

messages: Vec<Message>

Ordered list of messages in the conversation.

tools: Vec<ToolSpec>

Tool definitions available to the model.

tool_choice: ToolChoice

How the model should decide whether to call a tool.

response_format: Option<JsonSchema>

When set, instructs the provider to return structured JSON output.

max_tokens: Option<u32>

Maximum number of tokens the model may generate.

temperature: Option<f32>

Sampling temperature in [0.0, 2.0]. Lower is more deterministic.

stop: Vec<String>

Token sequences that cause the model to stop generating.

web_search: bool

When true, the model may search the public web to ground its answer.

This is a provider-agnostic capability hint: a provider maps it to its native mechanism (Vertex Gemini → the googleSearch grounding tool, alongside any tools function declarations) and a provider without web search ignores it. Defaults to false; the agent’s answering loop (run_turn) sets it from its RunTurnOptions.web_search, so auxiliary calls (summarization, classification) that bypass that loop never offer search.

Implementations

impl CompletionRequest

pub fn new(model: impl Into<String>) -> Self

Creates a new request for the given model with sensible defaults: empty messages, tools, and stop lists; tool_choice set to ToolChoice::Auto; all optional fields None.

Trait Implementations

impl Clone for CompletionRequest

fn clone(&self) -> CompletionRequest

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CompletionRequest

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for CompletionRequest

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for CompletionRequest

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

where __S: Serializer,

Serialize this value into the given Serde serializer.