feat: integrate graph orchestration modes into UI

- Extend orchestrator factory to support 5 modes (simple, advanced, iterative, deep, auto)
- Add lazy import for create_graph_orchestrator to avoid circular dependencies
- Update Gradio UI dropdown to show all 5 modes with descriptions
- Update example queries to demonstrate different orchestration modes (simple, iterative, deep)

src/app.py

@@ -555,10 +555,16 @@ def create_demo() -> gr.Blocks:
         # Settings will use defaults from configure_orchestrator
         with gr.Row(visible=False):
             mode_radio = gr.Radio(
-                choices=["simple", "advanced"],
+                choices=["simple", "advanced", "iterative", "deep", "auto"],
                 value="simple",
                 label="Orchestrator Mode",
-                info="Simple: Linear | Advanced: Multi-Agent (HuggingFace Inference)",
+                info=(
+                    "Simple: Linear search-judge loop | "
+                    "Advanced: Multi-agent (OpenAI) | "
+                    "Iterative: Knowledge-gap driven | "
+                    "Deep: Parallel sections | "
+                    "Auto: Smart routing"
+                ),
             )
 
             # Hidden text components for model/provider (not dropdowns to avoid value mismatch)
@@ -602,14 +608,14 @@ def create_demo() -> gr.Blocks:
                     "",
                 ],
                 [
-                    "Is metformin effective for treating cancer?",
-                    "simple",
+                    "Is metformin effective for treating cancer? Investigate mechanism of action.",
+                    "iterative",
                     "Qwen/Qwen3-235B-A22B-Instruct-2507",
                     "",
                 ],
                 [
-                    "What medications show promise for Long COVID treatment?",
-                    "simple",
+                    "Create a comprehensive report on Long COVID treatments including clinical trials, mechanisms, and safety.",
+                    "deep",
                     "zai-org/GLM-4.5-Air",
                     "nebius",
                 ],

src/orchestrator_factory.py

@@ -28,11 +28,24 @@ def _get_magentic_orchestrator_class() -> Any:
         ) from e
 
 
+def _get_graph_orchestrator_factory() -> Any:
+    """Import create_graph_orchestrator lazily to avoid circular dependencies."""
+    try:
+        from src.orchestrator.graph_orchestrator import create_graph_orchestrator
+
+        return create_graph_orchestrator
+    except ImportError as e:
+        logger.error("Failed to import create_graph_orchestrator", error=str(e))
+        raise ValueError(
+            "Graph orchestrators require Pydantic Graph. Please check dependencies."
+        ) from e
+
+
 def create_orchestrator(
     search_handler: SearchHandlerProtocol | None = None,
     judge_handler: JudgeHandlerProtocol | None = None,
     config: OrchestratorConfig | None = None,
-    mode: Literal["simple", "magentic", "advanced"] | None = None,
+    mode: Literal["simple", "magentic", "advanced", "iterative", "deep", "auto"] | None = None,
 ) -> Any:
     """
     Create an orchestrator instance.
@@ -41,7 +54,12 @@ def create_orchestrator(
         search_handler: The search handler (required for simple mode)
         judge_handler: The judge handler (required for simple mode)
         config: Optional configuration
-        mode: "simple", "magentic", "advanced" or None (auto-detect)
+        mode: Orchestrator mode - "simple", "advanced", "iterative", "deep", "auto", or None (auto-detect)
+            - "simple": Linear search-judge loop (Free Tier)
+            - "advanced": Multi-agent coordination (Requires OpenAI)
+            - "iterative": Knowledge-gap-driven research (Free Tier)
+            - "deep": Parallel section-based research (Free Tier)
+            - "auto": Intelligent mode detection (Free Tier)
 
     Returns:
         Orchestrator instance
@@ -55,6 +73,16 @@ def create_orchestrator(
             max_rounds=config.max_iterations if config else 10,
         )
 
+    # Graph-based orchestrators (iterative, deep, auto)
+    if effective_mode in ("iterative", "deep", "auto"):
+        create_graph_orchestrator = _get_graph_orchestrator_factory()
+        return create_graph_orchestrator(
+            mode=effective_mode,  # type: ignore[arg-type]
+            max_iterations=config.max_iterations if config else 5,
+            max_time_minutes=10,
+            use_graph=True,
+        )
+
     # Simple mode requires handlers
     if search_handler is None or judge_handler is None:
         raise ValueError("Simple mode requires search_handler and judge_handler")
@@ -71,9 +99,11 @@ def _determine_mode(explicit_mode: str | None) -> str:
     if explicit_mode:
         if explicit_mode in ("magentic", "advanced"):
             return "advanced"
+        if explicit_mode in ("iterative", "deep", "auto"):
+            return explicit_mode
         return "simple"
 
-    # Auto-detect: advanced if paid API key available
+    # Auto-detect: advanced if paid API key available, otherwise simple
     if settings.has_openai_key:
         return "advanced"