litellm-rs 0.1.1

//! Voyage AI provider implementation
//!
//! This module provides Voyage AI API integration for embeddings.

use super::{BaseProvider, ModelPricing, Provider, ProviderError, ProviderType};
use crate::config::ProviderConfig;
use crate::core::models::{RequestContext, openai::*};
use crate::utils::error::Result;
use async_trait::async_trait;
use serde_json::json;
use std::collections::HashMap;
use tracing::{debug, info};

/// Voyage AI provider implementation
#[derive(Debug, Clone)]
pub struct VoyageProvider {
    /// Base provider functionality
    base: BaseProvider,
    /// Model pricing cache
    pricing_cache: HashMap<String, ModelPricing>,
}

impl VoyageProvider {
    /// Create a new Voyage AI provider
    pub async fn new(config: &ProviderConfig) -> Result<Self> {
        let base = BaseProvider::new(config)?;

        let base_url = config
            .base_url
            .clone()
            .unwrap_or_else(|| "https://api.voyageai.com".to_string());

        let provider = Self {
            base: BaseProvider { base_url, ..base },
            pricing_cache: Self::initialize_pricing_cache(),
        };

        info!(
            "Voyage AI provider '{}' initialized successfully",
            config.name
        );
        Ok(provider)
    }

    /// Initialize pricing cache with Voyage AI model prices
    fn initialize_pricing_cache() -> HashMap<String, ModelPricing> {
        let mut cache = HashMap::new();

        // Voyage embedding models
        cache.insert(
            "voyage-large-2".to_string(),
            ModelPricing {
                model: "voyage-large-2".to_string(),
                input_cost_per_1k: 0.00012,
                output_cost_per_1k: 0.0,
                currency: "USD".to_string(),
                updated_at: chrono::Utc::now(),
            },
        );

        cache.insert(
            "voyage-code-2".to_string(),
            ModelPricing {
                model: "voyage-code-2".to_string(),
                input_cost_per_1k: 0.00012,
                output_cost_per_1k: 0.0,
                currency: "USD".to_string(),
                updated_at: chrono::Utc::now(),
            },
        );

        cache.insert(
            "voyage-2".to_string(),
            ModelPricing {
                model: "voyage-2".to_string(),
                input_cost_per_1k: 0.0001,
                output_cost_per_1k: 0.0,
                currency: "USD".to_string(),
                updated_at: chrono::Utc::now(),
            },
        );

        cache.insert(
            "voyage-lite-02-instruct".to_string(),
            ModelPricing {
                model: "voyage-lite-02-instruct".to_string(),
                input_cost_per_1k: 0.00001,
                output_cost_per_1k: 0.0,
                currency: "USD".to_string(),
                updated_at: chrono::Utc::now(),
            },
        );

        cache
    }
}

#[async_trait]
impl Provider for VoyageProvider {
    fn name(&self) -> &str {
        &self.base.name
    }

    fn provider_type(&self) -> ProviderType {
        ProviderType::Custom("voyage".to_string())
    }

    async fn supports_model(&self, model: &str) -> bool {
        self.base.is_model_supported(model) || model.starts_with("voyage")
    }

    async fn supports_images(&self) -> bool {
        false
    }

    async fn supports_embeddings(&self) -> bool {
        true // Voyage AI specializes in embeddings
    }

    async fn supports_streaming(&self) -> bool {
        false
    }

    async fn list_models(&self) -> Result<Vec<Model>> {
        let known_models = vec![
            "voyage-large-2",
            "voyage-code-2",
            "voyage-2",
            "voyage-lite-02-instruct",
        ];

        let models = known_models
            .into_iter()
            .map(|model| Model {
                id: model.to_string(),
                object: "model".to_string(),
                created: chrono::Utc::now().timestamp() as u64,
                owned_by: "voyage".to_string(),
            })
            .collect();

        Ok(models)
    }

    async fn health_check(&self) -> Result<()> {
        debug!("Performing Voyage AI health check");
        Ok(())
    }

    async fn chat_completion(
        &self,
        _request: ChatCompletionRequest,
        _context: RequestContext,
    ) -> Result<ChatCompletionResponse> {
        Err(
            ProviderError::InvalidRequest("Chat completion not supported by Voyage AI".to_string())
                .into(),
        )
    }

    async fn completion(
        &self,
        _request: CompletionRequest,
        _context: RequestContext,
    ) -> Result<CompletionResponse> {
        Err(
            ProviderError::InvalidRequest("Text completion not supported by Voyage AI".to_string())
                .into(),
        )
    }

    async fn embedding(
        &self,
        request: EmbeddingRequest,
        _context: RequestContext,
    ) -> Result<EmbeddingResponse> {
        debug!("Voyage AI embedding for model: {}", request.model);

        let body = json!({
            "model": request.model,
            "input": request.input
        });

        let url = format!("{}/v1/embeddings", self.base.base_url);

        let response = self
            .base
            .client
            .post(&url)
            .header("Authorization", format!("Bearer {}", self.base.api_key))
            .header("Content-Type", "application/json")
            .json(&body)
            .send()
            .await
            .map_err(|e| ProviderError::Network(e.to_string()))?;

        if !response.status().is_success() {
            let status = response.status();
            let error_text = response.text().await.unwrap_or_default();

            return Err(match status.as_u16() {
                401 => ProviderError::Authentication(error_text),
                429 => ProviderError::RateLimit(error_text),
                404 => ProviderError::ModelNotFound(error_text),
                400 => ProviderError::InvalidRequest(error_text),
                _ => ProviderError::Unknown(format!("HTTP {}: {}", status, error_text)),
            }
            .into());
        }

        let response_json: EmbeddingResponse = self.base.parse_json_response(response).await?;
        Ok(response_json)
    }

    async fn image_generation(
        &self,
        _request: ImageGenerationRequest,
        _context: RequestContext,
    ) -> Result<ImageGenerationResponse> {
        Err(ProviderError::InvalidRequest(
            "Image generation not supported by Voyage AI".to_string(),
        )
        .into())
    }

    async fn get_model_pricing(&self, model: &str) -> Result<ModelPricing> {
        if let Some(pricing) = self.pricing_cache.get(model) {
            Ok(pricing.clone())
        } else {
            Ok(ModelPricing {
                model: model.to_string(),
                input_cost_per_1k: 0.0001,
                output_cost_per_1k: 0.0,
                currency: "USD".to_string(),
                updated_at: chrono::Utc::now(),
            })
        }
    }

    async fn calculate_cost(
        &self,
        model: &str,
        input_tokens: u32,
        _output_tokens: u32,
    ) -> Result<f64> {
        let pricing = self.get_model_pricing(model).await?;

        // Embeddings only have input cost
        let input_cost = (input_tokens as f64 / 1000.0) * pricing.input_cost_per_1k;

        Ok(input_cost)
    }
}