Struct LlmClient 

pub struct LlmClient { /* private fields */ }

LLM client supporting both OpenAI and Claude API formats.

Implementations

impl LlmClient

pub fn new(api_base: &str, api_key: &str) -> Result<Self>

pub async fn chat_completion( &self, model: &str, messages: &[ChatMessage], tools: Option<&[ToolDefinition]>, temperature: Option<f64>, ) -> Result<ChatCompletionResponse>

Non-streaming chat completion call (auto-routes based on model/api_base).

pub async fn chat_completion_stream( &self, model: &str, messages: &[ChatMessage], tools: Option<&[ToolDefinition]>, temperature: Option<f64>, event_sink: &mut dyn EventSink, ) -> Result<ChatCompletionResponse>

Streaming chat completion call (auto-routes based on model/api_base).

pub async fn embed( &self, model: &str, texts: &[&str], custom_url: Option<&str>, custom_key: Option<&str>, ) -> Result<Vec<Vec<f32>>>

Embed text(s) using OpenAI-compatible /embeddings API. Returns one embedding vector per input string. Used when memory_vector feature is enabled. If custom_url and custom_key are provided, use them instead of self.api_base/self.api_key.

Handles both OpenAI-standard format ({"data": [{"embedding": [...]}]}) and Dashscope native format ({"output": {"embeddings": [{"embedding": [...]}]}}).