diff --git a/core/src/browser/extensions/engines/AIEngine.ts b/core/src/browser/extensions/engines/AIEngine.ts
index 2d75dea8f..7b08a455e 100644
--- a/core/src/browser/extensions/engines/AIEngine.ts
+++ b/core/src/browser/extensions/engines/AIEngine.ts
@@ -161,7 +161,7 @@ export interface modelInfo {
 export type listResult = modelInfo[]
 
 export interface SessionInfo {
-  pid: string // opaque handle for unload/chat
+  pid: number // opaque handle for unload/chat
   port: number // llama-server output port (corrected from portid)
   model_id: string, //name of the model
   model_path: string // path of the loaded model
diff --git a/extensions/llamacpp-extension/src/index.ts b/extensions/llamacpp-extension/src/index.ts
index 8293b420f..ff03d36d2 100644
--- a/extensions/llamacpp-extension/src/index.ts
+++ b/extensions/llamacpp-extension/src/index.ts
@@ -114,7 +114,7 @@ export default class llamacpp_extension extends AIEngine {
   readonly providerId: string = 'llamacpp'
 
   private config: LlamacppConfig
-  private activeSessions: Map<string, SessionInfo> = new Map()
+  private activeSessions: Map<number, SessionInfo> = new Map()
   private providerPath!: string
   private apiSecret: string = 'Jan'
 
@@ -724,7 +724,7 @@ export default class llamacpp_extension extends AIEngine {
       } catch (e) {}
       await this.sleep(500) // 500 sec interval during rechecks
     }
-    await this.unload(sInfo.pid)
+    await this.unload(sInfo.model_id)
     throw new Error(
       `Timed out loading model after ${timeoutMs}... killing llamacpp`
     )
@@ -967,10 +967,12 @@ export default class llamacpp_extension extends AIEngine {
     if (!sessionInfo) {
       throw new Error(`No active session found for model: ${opts.model}`)
     }
-    const result = invoke<boolean>('is_process_running', {
+    // check if the process is alive
+    const result = await invoke<boolean>('is_process_running', {
       pid: sessionInfo.pid,
     })
     if (!result) {
+      this.activeSessions.delete(sessionInfo.pid)
       throw new Error('Model have crashed! Please reload!')
     }
     const baseUrl = `http://localhost:${sessionInfo.port}/v1`
diff --git a/src-tauri/src/core/state.rs b/src-tauri/src/core/state.rs
index 6790dbb33..12cc34d4a 100644
--- a/src-tauri/src/core/state.rs
+++ b/src-tauri/src/core/state.rs
@@ -24,7 +24,7 @@ pub struct AppState {
     pub mcp_active_servers: Arc<Mutex<HashMap<String, serde_json::Value>>>,
     pub mcp_successfully_connected: Arc<Mutex<HashMap<String, bool>>>,
     pub server_handle: Arc<Mutex<Option<ServerHandle>>>,
-    pub llama_server_process: Arc<Mutex<HashMap<String, LLamaBackendSession>>>,
+    pub llama_server_process: Arc<Mutex<HashMap<i32, LLamaBackendSession>>>,
 }
 pub fn generate_app_token() -> String {
     rand::thread_rng()
diff --git a/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/cleanup.rs b/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/cleanup.rs
index 332c0c5a4..5e5a08fc5 100644
--- a/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/cleanup.rs
+++ b/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/cleanup.rs
@@ -3,7 +3,7 @@ use crate::core::state::AppState;
 
 pub async fn cleanup_processes(state: State<'_, AppState>) {
     let mut map = state.llama_server_process.lock().await;
-    let pids: Vec<String> = map.keys().cloned().collect();
+    let pids: Vec<i32> = map.keys().cloned().collect();
     for pid in pids {
         if let Some(session) = map.remove(&pid) {
             let mut child = session.child;
diff --git a/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/server.rs b/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/server.rs
index 7e84a5dad..adc0e1399 100644
--- a/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/server.rs
+++ b/src-tauri/src/core/utils/extensions/inference_llamacpp_extension/server.rs
@@ -9,7 +9,6 @@ use tauri::State; // Import Manager trait
 use thiserror;
 use tokio::process::Command;
 use tokio::time::timeout;
-use uuid::Uuid;
 
 use crate::core::state::AppState;
 use crate::core::state::LLamaBackendSession;
@@ -44,8 +43,8 @@ type ServerResult<T> = Result<T, ServerError>;
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct SessionInfo {
-    pub pid: String,  // opaque handle for unload/chat
-    pub port: String, // llama-server output port
+    pub pid: i32,  // opaque handle for unload/chat
+    pub port: i32, // llama-server output port
     pub model_id: String,
     pub model_path: String, // path of the loaded model
     pub api_key: String,
@@ -82,12 +81,19 @@ pub async fn load_llama_model(
         )));
     }
 
-    let port = args
+    let port_str = args
         .iter()
         .position(|arg| arg == "--port")
         .and_then(|i| args.get(i + 1))
         .cloned()
         .unwrap_or_default();
+    let port: i32 = match port_str.parse() {
+        Ok(p) => p,
+        Err(_) => {
+            eprintln!("Invalid port value: '{}', using default 8080", port_str);
+            8080
+        }
+    };
 
     let model_path = args
         .iter()
@@ -146,10 +152,7 @@ pub async fn load_llama_model(
     let child = command.spawn().map_err(ServerError::Io)?;
 
     // Get the PID to use as session ID
-    let pid = child.id().map(|id| id.to_string()).unwrap_or_else(|| {
-        // Fallback in case we can't get the PID for some reason
-        format!("unknown_pid_{}", Uuid::new_v4())
-    });
+    let pid = child.id().map(|id| id as i32).unwrap_or(-1);
 
     log::info!("Server process started with PID: {}", pid);
     let session_info = SessionInfo {
@@ -175,7 +178,7 @@ pub async fn load_llama_model(
 // --- Unload Command ---
 #[tauri::command]
 pub async fn unload_llama_model(
-    pid: String,
+    pid: i32,
     state: State<'_, AppState>,
 ) -> ServerResult<UnloadResult> {
     let mut map = state.llama_server_process.lock().await;
@@ -212,9 +215,7 @@ pub async fn unload_llama_model(
 
             if let Some(raw_pid) = child.id() {
                 log::info!("Sending Ctrl-C to PID {}", raw_pid);
-                let ok: i32 = unsafe {
-                    GenerateConsoleCtrlEvent(CTRL_C_EVENT, raw_pid as u32)
-                };
+                let ok: i32 = unsafe { GenerateConsoleCtrlEvent(CTRL_C_EVENT, raw_pid as u32) };
                 if ok == 0 {
                     log::error!("Failed to send Ctrl-C to PID {}", raw_pid);
                 }
@@ -266,9 +267,17 @@ pub fn generate_api_key(model_id: String, api_secret: String) -> Result<String,
 
 // process aliveness check
 #[tauri::command]
-pub fn is_process_running(pid: u32) -> Result<bool, String> {
+pub async fn is_process_running(pid: i32, state: State<'_, AppState>) -> Result<bool, String> {
     let mut system = System::new();
     system.refresh_processes(ProcessesToUpdate::All, true);
     let process_pid = Pid::from(pid as usize);
-    Ok(system.process(process_pid).is_some())
+    let alive = system.process(process_pid).is_some();
+
+    if !alive {
+        let mut map = state.llama_server_process.lock().await;
+        map.remove(&pid);
+    }
+
+    Ok(alive)
 }
+
diff --git a/src-tauri/tauri.conf.json b/src-tauri/tauri.conf.json
index 26c9629e7..fbd131ffa 100644
--- a/src-tauri/tauri.conf.json
+++ b/src-tauri/tauri.conf.json
@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Jan",
-  "version": "0.5.16",
+  "version": "0.6.900",
   "identifier": "jan.ai.app",
   "build": {
     "frontendDist": "../web-app/dist",