Struct InferenceRequest 

pub struct InferenceRequest {
    pub messages: Vec<Message>,
    pub system_prompt: Option<String>,
    pub model: Option<String>,
}

One completion request.

system-prompt and model are genuinely optional — none means “the host decides”, not “empty string”.

Fields

messages: Vec<Message>

Conversation to complete. Same {role, content} shape the compaction-event hands a hook, so a compaction hook can forward (a summarized slice of) what it was given without translation.

system_prompt: Option<String>

System prompt for this call. none sends no system prompt at all.

model: Option<String>

Model to run with. none resolves to the manifest’s primary inference.model.

A some(m) the driver or provider rejects returns err — the host never silently falls back to the primary model. A caller that wants fallback-on-failure must call run-inference a second time with model: none, so each attempt produces its own truthful trace record and OTel span rather than one span quietly relabelled.

Trait Implementations

impl Clone for InferenceRequest

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl ComponentType for InferenceRequest

impl Debug for InferenceRequest

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

Formats the value using the given formatter.

impl Lift for InferenceRequest

impl Lower for InferenceRequest