feat installer and download model on first launch

gui.py

@@ -408,6 +408,23 @@ def _insert_message_card(cursor: QTextCursor, accent: str, bg: str) -> QTextCurs
     return body_cell.firstCursorPosition()
 
 
+class ModelDownloadThread(QThread):
+    # qint64, not int - int maps to a 32-bit C int for this signal's
+    # cross-thread queued connection, which silently wraps a ~15GB byte
+    # count (e.g. 14_977_484_704 truncates to 2_092_582_816).
+    progress = pyqtSignal("qint64", "qint64")
+    done = pyqtSignal()
+    failed = pyqtSignal(str)
+
+    def run(self) -> None:
+        try:
+            chatbot.download_model(on_progress=self.progress.emit)
+        except Exception as e:
+            self.failed.emit(str(e))
+            return
+        self.done.emit()
+
+
 class ModelLoaderThread(QThread):
     loaded = pyqtSignal(object)
     failed = pyqtSignal(str)
@@ -586,9 +603,44 @@ class ChatWindow(QMainWindow):
 
         self._init_gpu_monitor()
 
-        self.statusBar().showMessage("Loading model...")
         self._set_input_enabled(False)
+        self._download_thread: ModelDownloadThread | None = None
+        self._loader_thread: ModelLoaderThread | None = None
 
+        if chatbot.MODEL_PATH.exists():
+            self._start_model_load()
+        else:
+            self._start_model_download()
+
+    def _start_model_download(self) -> None:
+        self.statusBar().showMessage("Downloading model (first run only)... 0%")
+        self._append_system(
+            "Downloading the language model for first-time setup. This is a "
+            "one-time, multi-gigabyte download and may take a while."
+        )
+        self._download_thread = ModelDownloadThread()
+        self._download_thread.progress.connect(self._on_download_progress)
+        self._download_thread.done.connect(self._on_download_done)
+        self._download_thread.failed.connect(self._on_model_failed)
+        self._download_thread.start()
+
+    def _on_download_progress(self, downloaded: int, total: int) -> None:
+        downloaded_gb = downloaded / (1024**3)
+        if total:
+            pct = 100 * downloaded / total
+            total_gb = total / (1024**3)
+            self.statusBar().showMessage(
+                f"Downloading model... {downloaded_gb:.2f}/{total_gb:.2f} GB ({pct:.0f}%)"
+            )
+        else:
+            self.statusBar().showMessage(f"Downloading model... {downloaded_gb:.2f} GB")
+
+    def _on_download_done(self) -> None:
+        self._download_thread = None
+        self._start_model_load()
+
+    def _start_model_load(self) -> None:
+        self.statusBar().showMessage("Loading model...")
         self._loader_thread = ModelLoaderThread()
         self._loader_thread.loaded.connect(self._on_model_loaded)
         self._loader_thread.failed.connect(self._on_model_failed)
@@ -995,7 +1047,7 @@ class ChatWindow(QMainWindow):
         # llama.cpp calls run in native code and can't be cancelled mid-call,
         # so wait for any in-flight load/generation to finish rather than
         # tearing down the thread while it's still running (which crashes Qt).
-        for thread in (self._loader_thread, self._ask_thread):
+        for thread in (self._download_thread, self._loader_thread, self._ask_thread):
             if thread is not None and thread.isRunning():
                 thread.wait()
         if self._gpu_handle is not None: