chat-api/src/dto/api.rs

use serde::{Deserialize, Serialize};
use utoipa::ToSchema;

#[derive(Debug, Serialize, ToSchema)]
pub struct ErrorResponse {
    pub error: String,
}

impl ErrorResponse {
    pub fn new(msg: impl Into<String>) -> Self {
        Self { error: msg.into() }
    }
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct ModelsResponse {
    pub models: Vec<ModelInfo>,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct ModelInfo {
    pub name: String,
    pub family: Option<String>,
    pub parameter_size: Option<String>,
    pub quantization: Option<String>,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct LoadModelResponse {
    pub model: String,
    pub status: String,
    pub keep_alive: String,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct LoadModelBody {
    pub keep_alive: Option<String>,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct UnloadModelResponse {
    pub model: String,
    pub status: String,
}

#[derive(Debug, Deserialize, Serialize, ToSchema, Default)]
pub struct BaseLLMRequest {
    pub model: String,

    #[serde(default)]
    pub stream: bool,

    pub temperature: Option<f32>,
    pub top_p: Option<f32>,

    // Ollama-native
    pub top_k: Option<u32>,
    pub repeat_penalty: Option<f32>,
    pub seed: Option<i64>,

    pub num_ctx: Option<u32>,
    pub num_predict: Option<u32>,

    pub stop: Option<Vec<String>>,

    pub keep_alive: Option<String>,
}

#[derive(Debug, Deserialize, Serialize, ToSchema)]
pub struct CompletionRequest {
    #[serde(flatten)]
    pub base: BaseLLMRequest,

    pub prompt: String,
}

#[derive(Debug, Serialize, Deserialize, ToSchema, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum CompletionObject {
    TextCompletion,
}

#[derive(Debug, Serialize, Deserialize, ToSchema, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum FinishReason {
    Stop,
    Length,
    ContentFilter,
    ToolCalls,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct CompletionResponse {
    pub id: String,
    pub object: CompletionObject,
    pub created: u64,
    pub model: String,
    pub choices: Vec<Choice>,
    pub usage: Usage,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct Choice {
    pub text: String,
    pub index: u32,
    pub finish_reason: FinishReason,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct Usage {
    pub prompt_tokens: u32,
    pub completion_tokens: u32,
    pub total_tokens: u32,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct CompletionChunk {
    pub id: String,
    pub object: String,
    pub choices: Vec<Choice>,
}

#[derive(Debug, Deserialize, Serialize, ToSchema)]
pub struct ChatRequest {
    #[serde(flatten)]
    pub base: BaseLLMRequest,

    pub messages: Vec<Message>,
}

#[derive(Debug, Deserialize, Serialize, ToSchema)]
pub struct Message {
    pub role: Role,
    pub content: String,
}

#[derive(Debug, Deserialize, Serialize, ToSchema, PartialEq, Eq)]
#[serde(rename_all = "lowercase")]
pub enum Role {
    System,
    User,
    Assistant,
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct ChatCompletionResponse {
    pub id: String,
    pub object: String,
    pub created: u64,
    pub model: String,
    pub choices: Vec<ChatChoice>,
    pub usage: Option<Usage>, // optional (Ollama may not always provide)
}

#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct ChatChoice {
    pub index: u32,
    pub message: Message,
    pub finish_reason: FinishReason,
}

#[derive(Debug, Serialize, ToSchema)]
pub struct ChatCompletionChunk {
    pub id: String,
    pub object: String,
    pub choices: Vec<ChatChunkChoice>,
}

#[derive(Debug, Serialize, ToSchema)]
pub struct ChatChunkChoice {
    pub index: u32,
    pub delta: ChatDelta,
    pub finish_reason: Option<FinishReason>,
}

#[derive(Debug, Serialize, ToSchema)]
pub struct ChatDelta {
    pub role: Option<Role>,
    pub content: Option<String>,
}

#[derive(serde::Deserialize)]
pub struct CreateApiKeyRequest {
    pub name: String,
    pub scopes: Vec<String>,
}

#[derive(serde::Serialize)]
pub struct CreateApiKeyResponse {
    pub api_key: String, // ONLY returned once
}