refactor: unified LLM config - CHAT_URL/VISION_URL/SUMMARY_URL with env var overrides

src/api/agent_api.rs

@@ -41,10 +41,10 @@ async fn translate_text(
         req.target_language, req.text
     );
 
-    // Call Gemma4 via llama.cpp (port 8082, OpenAI-compatible API)
+    // Call LLM via configurable endpoint
     let client = Client::new();
-    let llm_url = "http://localhost:8082/v1/chat/completions";
-    let model = "google_gemma-4-26B-A4B-it-Q5_K_M.gguf".to_string();
+    let llm_url = crate::core::config::llm::CHAT_URL.as_str();
+    let model = crate::core::config::llm::CHAT_MODEL.as_str();
 
     let body = serde_json::json!({
         "model": model,
@@ -71,20 +71,15 @@ async fn translate_text(
         )
     })?;
 
-    let translated_text = llm_resp
-        .get("choices")
-        .and_then(|c| c.as_array())
-        .and_then(|c| c.first())
-        .and_then(|c| c.get("message"))
-        .and_then(|m| m.get("content"))
-        .and_then(|v| v.as_str())
-        .unwrap_or("Translation failed")
+    let translated_text = llm_resp["choices"][0]["message"]["content"]
+        .as_str()
+        .unwrap_or("")
         .to_string();
 
     Ok(Json(TranslationResponse {
         success: true,
         translated_text,
-        source_language_detected: req.source_language.unwrap_or("unknown".to_string()),
-        model_used: model,
+        source_language_detected: req.source_language.unwrap_or_else(|| "auto".to_string()),
+        model_used: model.to_string(),
     }))
 }