Struct CompletionRequest 

pub struct CompletionRequest {
    pub model: Model,
    pub messages: Vec<Message>,
    pub tools: Vec<ToolDefinition>,
    pub max_tokens: Option<u32>,
    pub reasoning: ReasoningLevel,
    pub sampling: SamplingParams,
    pub modalities: Vec<Modality>,
    pub audio_config: Option<AudioOutputConfig>,
    pub image_config: Option<ImageGenConfig>,
    pub user: Option<String>,
    pub provider_preferences: Option<ProviderPreferences>,
    pub session_id: Option<SessionId>,
}

A single completion request sent to an LLM provider.

This is the primary input type for LlmProvider::complete. The agent runtime assembles this from the conversation history, active tools, and LLM config.

Fields

model: Model

Model to use (includes provider prefix).

messages: Vec<Message>

Message history in wire format.

tools: Vec<ToolDefinition>

Tools available to the model in this turn.

max_tokens: Option<u32>

Maximum output tokens. If None, the provider decides.

reasoning: ReasoningLevel

Whether reasoning/thinking mode is enabled.

sampling: SamplingParams

Sampling parameters (temperature, top_p, top_k, min_p, penalties).

modalities: Vec<Modality>

Output modalities (e.g. text+audio, image). Empty means text-only.

audio_config: Option<AudioOutputConfig>

Audio output configuration. Only used when modalities includes Modality::Audio.

image_config: Option<ImageGenConfig>

Image generation configuration. Only used when modalities includes Modality::Image.

user: Option<String>

Opaque user identifier for per-user tracking in provider dashboards.

provider_preferences: Option<ProviderPreferences>

Provider routing preferences (used by OpenRouter).

session_id: Option<SessionId>

Stable conversation identifier for provider cache routing.

Implementations

impl CompletionRequest

pub fn new(model: Model, messages: Vec<Message>) -> Self

Build a request with default settings and the supplied messages.

pub fn from_config(config: &AgentLlmConfig, messages: Vec<Message>) -> Self

Keep the agent’s session ID, routing, user and generation settings.

use flashmind_types::{AgentLlmConfig, CompletionRequest, Message, SessionId};

let session = SessionId::new();
let llm = AgentLlmConfig::new("openrouter:example/model".parse()?)
    .with_session_id(session);
let request = CompletionRequest::from_config(&llm, vec![Message::user("Hello")]);
assert_eq!(request.session_id, llm.session_id);

pub fn with_session_id(self, session_id: impl Into<SessionId>) -> Self

Use an existing session identity for this request.

Trait Implementations

impl Debug for CompletionRequest

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

Formats the value using the given formatter.