update: pipeline, search, clip, embedding fixes

src/api/agent_api.rs

@@ -41,22 +41,24 @@ async fn translate_text(
         req.target_language, req.text
     );
 
-    // Call Ollama API
+    // Call Gemma4 via llama.cpp (port 8082, OpenAI-compatible API)
     let client = Client::new();
-    let ollama_url = "http://localhost:11434/api/generate";
-
-    // Using qwen3:latest which is available locally
-    let model = "qwen3:latest".to_string();
+    let llm_url = "http://localhost:8082/v1/chat/completions";
+    let model = "google_gemma-4-26B-A4B-it-Q5_K_M.gguf".to_string();
 
     let body = serde_json::json!({
         "model": model,
-        "prompt": prompt,
-        "system": system_prompt,
-        "stream": false
+        "messages": [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": prompt}
+        ],
+        "stream": false,
+        "max_tokens": 1024,
+        "temperature": 0.1
     });
 
     let response = client
-        .post(ollama_url)
+        .post(llm_url)
         .json(&body)
         .send()
         .await
@@ -67,15 +69,19 @@ async fn translate_text(
             )
         })?;
 
-    let ollama_resp: serde_json::Value = response.json().await.map_err(|e| {
+    let llm_resp: serde_json::Value = response.json().await.map_err(|e| {
         (
             StatusCode::INTERNAL_SERVER_ERROR,
             format!("Failed to parse LLM response: {}", e),
         )
     })?;
 
-    let translated_text = ollama_resp
-        .get("response")
+    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")
         .to_string();