Use state for previous_response_id

Signed-off-by: Aivin V. Solatorio <avsolatorio@gmail.com>

mcp_openai_client.py

@@ -181,6 +181,7 @@ class MCPClientWrapper:
         history: List[Union[Dict[str, Any], ChatMessage]],
         previous_response_id: str = None,
     ):
+        print("previous_response_id", previous_response_id)
         if not self.session and LLM_PROVIDER == "anthropic":
             messages = history + [
                 {"role": "user", "content": message},
@@ -189,7 +190,7 @@ class MCPClientWrapper:
                     "content": "Please connect to an MCP server first by reloading the page.",
                 },
             ]
-            yield messages, gr.Textbox(value=""), gr.Textbox(value=previous_response_id)
+            yield messages, gr.Textbox(value=""), previous_response_id
         else:
             messages = history + [
                 {"role": "user", "content": message},
@@ -199,9 +200,9 @@ class MCPClientWrapper:
                 },
             ]
 
-            yield messages, gr.Textbox(value=""), gr.Textbox(value=previous_response_id)
+            yield messages, gr.Textbox(value=""), previous_response_id
             # simulate thinking with asyncio.sleep
-            await asyncio.sleep(0.1)
+            await asyncio.sleep(0.2)
             messages.pop(-1)
 
             is_delta = False
@@ -227,7 +228,7 @@ class MCPClientWrapper:
                         yield (
                             messages,
                             gr.Textbox(value=""),
-                            gr.Textbox(value=previous_response_id),
+                            previous_response_id,
                         )
                         await asyncio.sleep(0.01)
                         continue
@@ -238,7 +239,7 @@ class MCPClientWrapper:
                 yield (
                     messages,
                     gr.Textbox(value=""),
-                    gr.Textbox(value=previous_response_id),
+                    previous_response_id,
                 )
                 await asyncio.sleep(0.01)
 
@@ -250,7 +251,15 @@ class MCPClientWrapper:
                     break
 
         with open("messages.log.jsonl", "a+") as fl:
-            fl.write(json.dumps(dict(time=f"{datetime.now()}", messages=messages)))
+            fl.write(
+                json.dumps(
+                    dict(
+                        time=f"{datetime.now()}",
+                        messages=messages,
+                        previous_response_id=previous_response_id,
+                    )
+                )
+            )
 
     async def _process_query_openai(
         self,
@@ -278,8 +287,8 @@ class MCPClientWrapper:
             stream=True,
             max_output_tokens=32768,
             temperature=0,
-            previous_response_id=previous_response_id
-            if previous_response_id.strip()
+            previous_response_id=previous_response_id.strip()
+            if previous_response_id
             else None,
             store=True,  # Store the response in the OpenAIlogs
         )
@@ -700,9 +709,7 @@ def gradio_interface(
                 layout="panel",
                 placeholder="Ask development data questions!",
             )
-            previous_response_id = gr.Textbox(
-                label="Previous Response ID", interactive=False, visible=False
-            )
+            previous_response_id = gr.State(None)
 
             with gr.Row(equal_height=True):
                 msg = gr.Textbox(