feat: Add dual output support to Agent component with structured JSON parsing (#8836)

* feat: Add dual output support to Agent component with structured JSON parsing

## Summary
- Add "Structured Response" output alongside existing "Response" output
- Filter out conflicting json_mode field from OpenAI inputs
- Implement robust JSON parsing with fallback handling

## Changes Made
### Agent Component (agent.py)
- Add second output: "Structured Response" (Data type) with tool_mode=False
- Filter json_mode from OpenAI inputs to prevent UI conflicts
- Add json_response() method with multi-stage JSON parsing:
  - Direct JSON parsing for valid responses
  - Regex extraction for embedded JSON in text
  - Graceful error handling with diagnostic info
- Share execution between outputs (no duplicate agent runs)
- Fix model building to handle missing json_mode attribute

### Tests (test_agent_component.py)
- Add 9 comprehensive test cases covering:
  - Dual output structure validation
  - Input filtering verification
  - JSON parsing (valid, embedded, error cases)
  - Model building without json_mode
  - Shared execution efficiency
  - Frontend node structure
  - Component initialization

## Benefits
- Users get both Message and Data output types to choose from
- Clean UI without confusing duplicate JSON toggles
- Robust JSON parsing handles various response formats
- Efficient single-execution approach
- Maintains backward compatibility

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

* [autofix.ci] apply automated fixes

* update to templates with  model list update

* [autofix.ci] apply automated fixes

* Update test_agent_component.py

* update to the test and update to templates

* [autofix.ci] apply automated fixes

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: Edwin Jose <edwin.jose@datastax.com>

src/backend/base/langflow/base/models/openai_constants.py

@@ -8,7 +8,9 @@ OPENAI_MODELS_DETAILED = [
     create_model_metadata(provider="OpenAI", name="gpt-4.1", icon="OpenAI", tool_calling=True),
     create_model_metadata(provider="OpenAI", name="gpt-4.1-mini", icon="OpenAI", tool_calling=True),
     create_model_metadata(provider="OpenAI", name="gpt-4.1-nano", icon="OpenAI", tool_calling=True),
-    create_model_metadata(provider="OpenAI", name="gpt-4.5-preview", icon="OpenAI", tool_calling=True, preview=True),
+    create_model_metadata(
+        provider="OpenAI", name="gpt-4.5-preview", icon="OpenAI", tool_calling=True, preview=True, not_supported=True
+    ),
     create_model_metadata(provider="OpenAI", name="gpt-4-turbo", icon="OpenAI", tool_calling=True),
     create_model_metadata(
         provider="OpenAI", name="gpt-4-turbo-preview", icon="OpenAI", tool_calling=True, preview=True
@@ -17,8 +19,8 @@ OPENAI_MODELS_DETAILED = [
     create_model_metadata(provider="OpenAI", name="gpt-3.5-turbo", icon="OpenAI", tool_calling=True),
     # Reasoning Models
     create_model_metadata(provider="OpenAI", name="o1", icon="OpenAI", reasoning=True),
-    create_model_metadata(provider="OpenAI", name="o1-mini", icon="OpenAI", reasoning=True),
-    create_model_metadata(provider="OpenAI", name="o1-pro", icon="OpenAI", reasoning=True),
+    create_model_metadata(provider="OpenAI", name="o1-mini", icon="OpenAI", reasoning=True, not_supported=True),
+    create_model_metadata(provider="OpenAI", name="o1-pro", icon="OpenAI", reasoning=True, not_supported=True),
     create_model_metadata(provider="OpenAI", name="o3-mini", icon="OpenAI", reasoning=True),
     create_model_metadata(provider="OpenAI", name="o3", icon="OpenAI", reasoning=True),
     create_model_metadata(provider="OpenAI", name="o3-pro", icon="OpenAI", reasoning=True),

src/backend/base/langflow/components/agents/agent.py

@@ -1,3 +1,6 @@
+import json
+import re
+
 from langchain_core.tools import StructuredTool
 
 from langflow.base.agents.agent import LCToolsAgentComponent
@@ -18,6 +21,7 @@ from langflow.custom.utils import update_component_build_config
 from langflow.field_typing import Tool
 from langflow.io import BoolInput, DropdownInput, IntInput, MultilineInput, Output
 from langflow.logging import logger
+from langflow.schema.data import Data
 from langflow.schema.dotdict import dotdict
 from langflow.schema.message import Message
 
@@ -40,6 +44,13 @@ class AgentComponent(ToolCallingAgentComponent):
 
     memory_inputs = [set_advanced_true(component_input) for component_input in MemoryComponent().inputs]
 
+    # Filter out json_mode from OpenAI inputs since we handle structured output differently
+    openai_inputs_filtered = [
+        input_field
+        for input_field in MODEL_PROVIDERS_DICT["OpenAI"]["inputs"]
+        if not (hasattr(input_field, "name") and input_field.name == "json_mode")
+    ]
+
     inputs = [
         DropdownInput(
             name="agent_llm",
@@ -51,7 +62,7 @@ class AgentComponent(ToolCallingAgentComponent):
             input_types=[],
             options_metadata=[MODELS_METADATA[key] for key in MODEL_PROVIDERS_LIST] + [{"icon": "brain"}],
         ),
-        *MODEL_PROVIDERS_DICT["OpenAI"]["inputs"],
+        *openai_inputs_filtered,
         MultilineInput(
             name="system_prompt",
             display_name="Agent Instructions",
@@ -78,7 +89,10 @@ class AgentComponent(ToolCallingAgentComponent):
             value=True,
         ),
     ]
-    outputs = [Output(name="response", display_name="Response", method="message_response")]
+    outputs = [
+        Output(name="response", display_name="Response", method="message_response"),
+        Output(name="structured_response", display_name="Structured Response", method="json_response", tool_mode=False),
+    ]
 
     async def message_response(self) -> Message:
         try:
@@ -114,7 +128,11 @@ class AgentComponent(ToolCallingAgentComponent):
                 system_prompt=self.system_prompt,
             )
             agent = self.create_agent_runnable()
-            return await self.run_agent(agent)
+            result = await self.run_agent(agent)
+
+            # Store result for potential JSON output
+            self._agent_result = result
+            # return result
 
         except (ValueError, TypeError, KeyError) as e:
             logger.error(f"{type(e).__name__}: {e!s}")
@@ -125,6 +143,41 @@ class AgentComponent(ToolCallingAgentComponent):
         except Exception as e:
             logger.error(f"Unexpected error: {e!s}")
             raise
+        else:
+            return result
+
+    async def json_response(self) -> Data:
+        """Convert agent response to structured JSON Data output."""
+        # Run the regular message response first to get the result
+        if not hasattr(self, "_agent_result"):
+            await self.message_response()
+
+        result = self._agent_result
+
+        # Extract content from result
+        if hasattr(result, "content"):
+            content = result.content
+        elif hasattr(result, "text"):
+            content = result.text
+        else:
+            content = str(result)
+
+        # Try to parse as JSON
+        try:
+            json_data = json.loads(content)
+            return Data(data=json_data)
+        except json.JSONDecodeError:
+            # If it's not valid JSON, try to extract JSON from the content
+            json_match = re.search(r"\{.*\}", content, re.DOTALL)
+            if json_match:
+                try:
+                    json_data = json.loads(json_match.group())
+                    return Data(data=json_data)
+                except json.JSONDecodeError:
+                    pass
+
+            # If we can't extract JSON, return the raw content as data
+            return Data(data={"content": content, "error": "Could not parse as JSON"})
 
     async def get_memory_data(self):
         # TODO: This is a temporary fix to avoid message duplication. We should develop a function for this.
@@ -171,7 +224,11 @@ class AgentComponent(ToolCallingAgentComponent):
         if provider_info:
             inputs = provider_info.get("inputs")
             prefix = provider_info.get("prefix")
-            model_kwargs = {input_.name: getattr(self, f"{prefix}{input_.name}") for input_ in inputs}
+            # Filter out json_mode and only use attributes that exist on this component
+            model_kwargs = {}
+            for input_ in inputs:
+                if hasattr(self, f"{prefix}{input_.name}"):
+                    model_kwargs[input_.name] = getattr(self, f"{prefix}{input_.name}")
 
             return component.set(**model_kwargs)
         return component