feat(chatbot): improve stability, model handling and testing

run_chatbot.py

@@ -0,0 +1,98 @@
+"""
+run_chatbot.py — Standalone launcher for the eSim AI Chatbot
+=====================================================================
+Run from the project root:
+    python run_chatbot.py
+
+Requirements:
+    pip install PyQt5 ollama pillow
+
+For voice input:
+    pip install SpeechRecognition pyaudio
+
+For offline voice (no internet):
+    pip install faster-whisper
+
+LLM backend:
+    1. Install Ollama from https://ollama.com
+    2. ollama serve
+    3. ollama pull qwen2.5:3b     # text model (~2GB)
+    4. ollama pull moondream      # vision model (~1.6GB, optional)
+"""
+import os
+import sys
+
+# Add src/ to Python path so all local imports resolve correctly
+_HERE = os.path.dirname(os.path.abspath(__file__))
+_SRC  = os.path.join(_HERE, 'src')
+if _SRC not in sys.path:
+    sys.path.insert(0, _SRC)
+
+# Minimal Appconfig stub so Chatbot.py imports cleanly without the
+# full eSim configuration layer (which expects an installed eSim).
+# Only used when running chatbot standalone (not via Application.py).
+import types
+_conf_mod = types.ModuleType("configuration")
+_app_mod  = types.ModuleType("configuration.Appconfig")
+
+class _AppConfigStub:
+    """Minimal stub — only attributes that Chatbot.py actually uses."""
+    noteArea = {}
+    procThread_list = []
+    process_obj = []
+    _APPLICATION = "eSim"
+    _VERSION = "2.4"
+    current_project = {"ProjectName": None}
+    def print_info(self, *a): pass
+    def print_error(self, *a): pass
+
+_app_mod.Appconfig = _AppConfigStub
+_conf_mod.Appconfig = _app_mod
+sys.modules["configuration"] = _conf_mod
+sys.modules["configuration.Appconfig"] = _app_mod
+
+# Now safe to import the chatbot UI
+from PyQt5.QtWidgets import QApplication
+from PyQt5.QtGui import QFont
+from frontEnd.Chatbot import ChatbotGUI  # noqa: E402
+
+
+def main():
+    app = QApplication(sys.argv)
+    app.setApplicationName("eSim AI Chatbot")
+    app.setFont(QFont("Segoe UI", 10))
+
+    # High-DPI support (Windows)
+    try:
+        from PyQt5.QtCore import Qt
+        QApplication.setAttribute(Qt.AA_EnableHighDpiScaling, True)
+        QApplication.setAttribute(Qt.AA_UseHighDpiPixmaps, True)
+    except Exception:
+        pass
+
+    window = ChatbotGUI()
+    window.setWindowTitle("eSim AI Assistant — Standalone")
+    window.resize(520, 640)
+    window.show()
+
+    print("=" * 60)
+    print("  eSim AI Chatbot — running standalone")
+    print("=" * 60)
+    print()
+    print("  Status: Window opened [OK]")
+    print()
+    print("  To enable AI responses:")
+    print("    1. Install Ollama: https://ollama.com")
+    print("    2. Run:  ollama serve")
+    print("    3. Run:  ollama pull qwen2.5:3b")
+    print("    4. Restart this script")
+    print()
+    print("  The Live/Offline indicator in the top-right")
+    print("  of the chat window shows Ollama's status.")
+    print("=" * 60)
+
+    sys.exit(app.exec_())
+
+
+if __name__ == "__main__":
+    main()

src/chatbot/chatbot_thread.py

@@ -62,9 +62,29 @@ def _downscale_image_bytes(raw_bytes: bytes) -> bytes:
 
 
 def get_stt_backend() -> str:
-    """Returns 'google' if SpeechRecognition is installed, else 'none'."""
+    """Return the best available speech-to-text backend.
+
+    Priority: faster-whisper (offline, best quality) > vosk (offline, lightweight)
+    > google (online, no install) > none
+    """
+    # Check faster-whisper first: offline, no internet needed
+    try:
+        import faster_whisper  # noqa: F401
+        return "whisper"
+    except ImportError:
+        pass
+
+    # Check vosk: offline, very lightweight
+    try:
+        import vosk  # noqa: F401
+        return "vosk"
+    except ImportError:
+        pass
+
+    # Fall back to Google online STT (requires internet)
     if _SR_AVAILABLE:
         return "google"
+
     return "none"
 
 
@@ -85,13 +105,19 @@ def start_ollama(stop_flag=None):
     The polling loop checks stop_flag() each second and exits early if cancelled.
     """
     if os.name == 'nt':
-        subprocess.Popen('start cmd /k "ollama serve"', shell=True)
+        # Run ollama serve silently in the background without a visible CMD window.
+        # CREATE_NO_WINDOW prevents a terminal from popping up and staying open.
+        subprocess.Popen(
+            ['ollama', 'serve'],
+            creationflags=subprocess.CREATE_NO_WINDOW,
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
+        )
     else:
         subprocess.Popen(
-            ['bash', '-c',
-             'x-terminal-emulator -e "ollama serve" || '
-             'gnome-terminal -- ollama serve || '
-             'xterm -e "ollama serve"']
+            ['bash', '-c', 'ollama serve'],
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
         )
 
     for _ in range(30):
@@ -344,9 +370,10 @@ def run(self):
             bot_response = ""
             for chunk in stream:
                 if self._stop_requested:
-                    bot_response += "\n\n⏹ Generation stopped."
+                    bot_response += "\n\n\u23f9 Generation stopped."
                     break
-                bot_response += chunk["message"]["content"]
+                # Use .get() to safely handle malformed/done chunks from Ollama
+                bot_response += chunk.get("message", {}).get("content", "")
 
             bot_response = bot_response.strip()
             if not bot_response:
@@ -510,7 +537,8 @@ def _chat_once(self, model_name: str, prompt: str, image_bytes_list):
             if self._stop_requested:
                 response += "\n\n\u23f9 Generation stopped."
                 break
-            piece       = chunk["message"]["content"]
+            # Use .get() to safely handle malformed/done chunks from Ollama
+            piece       = chunk.get("message", {}).get("content", "")
             response   += piece
             token_count += 1