trace any

app.py

@@ -2,20 +2,15 @@ import os
 import re
 import streamlit as st
 from dotenv import load_dotenv
-from langchain.agents.openai_assistant import OpenAIAssistantRunnable
+import openai
+from langsmith import traceable
 
 # Load environment variables
 load_dotenv()
 api_key = os.getenv("OPENAI_API_KEY")
-extractor_agent = os.getenv("ASSISTANT_ID_SOLUTION_SPECIFIER_A")
-
-# Create the assistant
-extractor_llm = OpenAIAssistantRunnable(
-    assistant_id=extractor_agent,
-    api_key=api_key,
-    as_agent=True
-)
+openai.api_key = api_key
 
+# Helper function to remove citations
 def remove_citation(text: str) -> str:
     pattern = r"【\d+†\w+】"
     return re.sub(pattern, "📚", text)
@@ -28,21 +23,36 @@ if "thread_id" not in st.session_state:
 
 st.title("Solution Specifier A")
 
-def predict(user_input: str) -> str:
+# Traceable function for predict logic
+@traceable
+def get_response(user_input: str, thread_id: str = None):
     """
-    This function calls our OpenAIAssistantRunnable to get a response.
-    If we don't have a thread_id yet, we create a new thread on the first call.
-    Otherwise, we continue the existing thread.
+    This function calls OpenAI API to get a response.
+    If thread_id is provided, it continues the conversation.
+    Otherwise, it starts a new conversation.
     """
-    if st.session_state["thread_id"] is None:
-        response = extractor_llm.invoke({"content": user_input})
-        st.session_state["thread_id"] = response.thread_id
+    messages = [{"role": "user", "content": user_input}]
+    if thread_id:
+        response = openai.ChatCompletion.create(
+            model="gpt-3.5-turbo",
+            messages=messages,
+            user=thread_id
+        )
     else:
-        response = extractor_llm.invoke(
-            {"content": user_input, "thread_id": st.session_state["thread_id"]}
+        response = openai.ChatCompletion.create(
+            model="gpt-3.5-turbo",
+            messages=messages
         )
-    output = response.return_values["output"]
-    return remove_citation(output)
+    return response["choices"][0]["message"]["content"], response["id"]
+
+# Streamlit app logic
+def predict(user_input: str) -> str:
+    if st.session_state["thread_id"] is None:
+        response_text, thread_id = get_response(user_input)
+        st.session_state["thread_id"] = thread_id
+    else:
+        response_text, _ = get_response(user_input, thread_id=st.session_state["thread_id"])
+    return remove_citation(response_text)
 
 # Display any existing messages (from a previous run or refresh)
 for msg in st.session_state["messages"]:

requirements.txt

@@ -1,3 +1,4 @@
 langchain
 langchain-openai
-python-dotenv
+python-dotenv
+langsmith