make openai optional in tui and lazy client creation in backend

src/config/settings.py

@@ -279,7 +279,7 @@ class AppClients:
         self.opensearch = None
         self.langflow_client = None
         self.langflow_http_client = None
-        self.patched_async_client = None
+        self._patched_async_client = None  # Private attribute
         self.converter = None
 
     async def initialize(self):
@@ -318,8 +318,21 @@ class AppClients:
                 "No Langflow client initialized yet, will attempt later on first use"
             )
 
-        # Initialize patched OpenAI client
+        # Initialize patched OpenAI client if API key is available
-        self.patched_async_client = patch_openai_with_mcp(AsyncOpenAI())
+        # This allows the app to start even if OPENAI_API_KEY is not set yet
+        # (e.g., when it will be provided during onboarding)
+        # The property will handle lazy initialization if needed later
+        if self._patched_async_client is None:
+            try:
+                openai_key = os.getenv("OPENAI_API_KEY")
+                if openai_key:
+                    self._patched_async_client = patch_openai_with_mcp(AsyncOpenAI())
+                    logger.info("OpenAI client initialized with API key from environment")
+                else:
+                    logger.info("OpenAI API key not found in environment - will be initialized on first use if needed")
+            except Exception as e:
+                logger.warning("Failed to initialize OpenAI client", error=str(e))
+                self._patched_async_client = None
 
         # Initialize document converter
         self.converter = create_document_converter(ocr_engine=DOCLING_OCR_ENGINE)
@@ -350,6 +363,41 @@ class AppClients:
                 self.langflow_client = None
         return self.langflow_client
 
+    @property
+    def patched_async_client(self):
+        """
+        Property that ensures OpenAI client is initialized on first access.
+        This allows lazy initialization so the app can start without an API key.
+        """
+        if self._patched_async_client is not None:
+            return self._patched_async_client
+
+        # Try to initialize the client on-demand
+        # First check if OPENAI_API_KEY is in environment
+        openai_key = os.getenv("OPENAI_API_KEY")
+
+        if not openai_key:
+            # Try to get from config (in case it was set during onboarding)
+            try:
+                config = get_openrag_config()
+                if config and config.provider and config.provider.api_key:
+                    openai_key = config.provider.api_key
+                    # Set it in environment so AsyncOpenAI can pick it up
+                    os.environ["OPENAI_API_KEY"] = openai_key
+                    logger.info("Loaded OpenAI API key from config file")
+            except Exception as e:
+                logger.debug("Could not load OpenAI key from config", error=str(e))
+
+        # Try to initialize the client - AsyncOpenAI() will read from environment
+        try:
+            self._patched_async_client = patch_openai_with_mcp(AsyncOpenAI())
+            logger.info("OpenAI client initialized on-demand")
+        except Exception as e:
+            logger.error("Failed to initialize OpenAI client on-demand", error=str(e))
+            raise ValueError(f"Failed to initialize OpenAI client: {str(e)}. Please complete onboarding or set OPENAI_API_KEY environment variable.")
+
+        return self._patched_async_client
+
     async def langflow_request(self, method: str, endpoint: str, **kwargs):
         """Central method for all Langflow API requests"""
         api_key = await generate_langflow_api_key()

src/connectors/service.py

@@ -21,7 +21,7 @@ class ConnectorService:
         task_service=None,
         session_manager=None,
     ):
-        self.openai_client = patched_async_client
+        self.clients = patched_async_client  # Store the clients object to access the property
         self.process_pool = process_pool
         self.embed_model = embed_model
         self.index_name = index_name

src/main.py

@@ -470,7 +470,7 @@ async def initialize_services():
         session_manager=session_manager,
     )
     openrag_connector_service = ConnectorService(
-        patched_async_client=clients.patched_async_client,
+        patched_async_client=clients,  # Pass the clients object itself
         process_pool=process_pool,
         embed_model=get_embedding_model(),
         index_name=INDEX_NAME,

src/tui/managers/env_manager.py

@@ -170,8 +170,9 @@ class EnvManager:
         """
         self.config.validation_errors.clear()
 
-        # Always validate OpenAI API key
+        # OpenAI API key is now optional (can be provided during onboarding)
-        if not validate_openai_api_key(self.config.openai_api_key):
+        # Only validate format if a key is provided
+        if self.config.openai_api_key and not validate_openai_api_key(self.config.openai_api_key):
             self.config.validation_errors["openai_api_key"] = (
                 "Invalid OpenAI API key format (should start with sk-)"
             )
@@ -268,7 +269,9 @@ class EnvManager:
                 f.write(f"LANGFLOW_URL_INGEST_FLOW_ID={self._quote_env_value(self.config.langflow_url_ingest_flow_id)}\n")
                 f.write(f"NUDGES_FLOW_ID={self._quote_env_value(self.config.nudges_flow_id)}\n")
                 f.write(f"OPENSEARCH_PASSWORD={self._quote_env_value(self.config.opensearch_password)}\n")
-                f.write(f"OPENAI_API_KEY={self._quote_env_value(self.config.openai_api_key)}\n")
+                # Only write OpenAI API key if provided (can be set during onboarding instead)
+                if self.config.openai_api_key:
+                    f.write(f"OPENAI_API_KEY={self._quote_env_value(self.config.openai_api_key)}\n")
                 f.write(
                     f"OPENRAG_DOCUMENTS_PATHS={self._quote_env_value(self.config.openrag_documents_paths)}\n"
                 )
@@ -345,7 +348,7 @@ class EnvManager:
     def get_no_auth_setup_fields(self) -> List[tuple[str, str, str, bool]]:
         """Get fields required for no-auth setup mode. Returns (field_name, display_name, placeholder, can_generate)."""
         return [
-            ("openai_api_key", "OpenAI API Key", "sk-...", False),
+            ("openai_api_key", "OpenAI API Key", "sk-... or leave empty", False),
             (
                 "opensearch_password",
                 "OpenSearch Password",

src/tui/screens/config.py

@@ -203,12 +203,16 @@ class ConfigScreen(Screen):
         yield Static(" ")
 
         # OpenAI API Key
-        yield Label("OpenAI API Key *")
+        yield Label("OpenAI API Key")
         # Where to create OpenAI keys (helper above the box)
         yield Static(
             Text("Get a key: https://platform.openai.com/api-keys", style="dim"),
             classes="helper-text",
         )
+        yield Static(
+            Text("Can also be provided during onboarding", style="dim italic"),
+            classes="helper-text",
+        )
         current_value = getattr(self.env_manager.config, "openai_api_key", "")
         with Horizontal(id="openai-key-row"):
             input_widget = Input(