Refine LangGraphAgent message handling and adjust webpage extraction tool depth to basic

agents.py

@@ -37,7 +37,7 @@ rate_limiter = InMemoryRateLimiter(
 
 class LangGraphAgent:
     def __init__(self, model_name: str = "gpt-4o",):
-        self.llm = ChatOpenAI(model=model_name, max_tokens=3000, temperature=0, streaming=True, rate_limiter=rate_limiter)
+        self.llm = ChatOpenAI(model=model_name, max_tokens=2000, temperature=0, rate_limiter=rate_limiter)
         self.llm_with_tools = self.llm.bind_tools(tools)
         self.tool_node = ToolNode(tools)
         self.graph = self.create_graph().compile()
@@ -63,12 +63,17 @@ class LangGraphAgent:
         if len(messages) == 1:
             system_prompt = ("You are a helpful assistant that can answer questions using various tools. "
             "You must answer the given question using as few words as possible, or the given format, if any."
-            " If the answer is a number, you must return the number only, do not include symbols or commas. If you don't know the answer, say 'I don't know'. "
-            "These questions are nontrivial and may require advanced critical thinking and multiple tool calls."
-            "If you need to search the web for information, always use a search tool before using a wepbage extraction tool so you always have a legit website."
+            "If the answer is a number, you must return the number only, do not include symbols or commas."
+            "If you need to search the web for information and aren't given a URL, always use a search tool before using a wepbage extraction tool so you always have a legit website."
             "If given a Python file, execute it with the code interpreter tool (riza_exec_python)")
             messages = [SystemMessage(system_prompt)] + messages
         
+        MAX_HISTORY = 3  # tune as needed
+
+        # Keep system + last N messages
+        if len(messages) > MAX_HISTORY:
+            messages = [m for m in messages if isinstance(m, SystemMessage)] + messages[-MAX_HISTORY:]
+
         while True:
             response = self.llm_with_tools.invoke(messages)
             messages.append(response)

tools.py

@@ -11,21 +11,6 @@ import requests
 import base64
 import pandas as pd
 
-# def search_tool(query: str) -> str:
-#     """
-#     Search the web for information using Tavily API (or similar service).
-#     """
-#     api_key = os.environ.get("TAVILY_API_KEY")
-#     if not api_key:
-#         return "TAVILY_API_KEY not set."
-#     url = f"https://api.tavily.com/search?q={query}&key={api_key}"
-#     try:
-#         response = requests.get(url)
-#         response.raise_for_status()
-#         return response.text
-#     except Exception as e:
-#         return f"Error searching web: {e}"
-
 @tool
 def describe_image_tool(image_path: str, prompt: str = "Describe the following image:") -> str:
     """
@@ -69,58 +54,6 @@ def parse_excel_tool(file_path: str) -> str:
     except Exception as e:
         return f"Error parsing Excel file: {e}"
 
-@tool
-def access_webpage_tool(url: str) -> str:
-    """
-    Access a webpage and return its content. This should not be used for YouTube
-    Args:
-        url (str): URL of the webpage to access.
-    Returns:
-        str: Content of the webpage or an error message.
-    """
-    print(f"Accessing webpage: {url}")
-    try:
-        response = requests.get(url)
-        response.raise_for_status()
-        return response.text
-    except Exception as e:
-        return f"Error accessing {url}: {e}"
-# @tool
-# def string_functions_tool(input_string: str, operation: str) -> str:
-#     """
-#     Perform various string operations based on the specified operation.
-#     Args:
-#         input_string (str): The string to operate on.
-#         operation (str): The operation to perform. Supported operations: 
-#                          'uppercase', 'lowercase', 'reverse', 'length', 
-#                          'count_vowels', 'count_consonants', 'count_words', 
-#                          'count_sentences'.
-#     Returns:
-#         str: Result of the operation or an error message if the operation is invalid.
-#     """
-#     if operation == "uppercase":
-#         return input_string.upper()
-#     elif operation == "lowercase":
-#         return input_string.lower()
-#     elif operation == "reverse":
-#         return input_string[::-1]
-#     elif operation == "length":
-#         return str(len(input_string))
-#     elif operation == "count_vowels":
-#         vowels = "aeiouAEIOU"
-#         return str(sum(1 for char in input_string if char in vowels))
-#     elif operation == "count_consonants":
-#         vowels = "aeiouAEIOU"
-#         return str(sum(1 for char in input_string if char.isalpha() and char not in vowels))
-#     elif operation == "count_words":
-#         return str(len(input_string.split()))
-#     elif operation == "count_sentences":
-#         import re
-#         sentences = re.split(r'[.!?]+', input_string)
-#         return str(len([s for s in sentences if s.strip()]))
-#     else:
-#         return "Invalid operation. Supported operations: uppercase, lowercase, reverse."
-
 @tool
 def audio_file_transcriber(file_path: str) -> str :
     """
@@ -135,7 +68,7 @@ def audio_file_transcriber(file_path: str) -> str :
     return docs[0].page_content
 
 
-brave_web_search = BraveSearch.from_search_kwargs({"max_results": 5})
+brave_web_search = BraveSearch.from_search_kwargs({"max_results": 4})
 
 
 # search_tool = TavilySearch(
@@ -147,7 +80,7 @@ brave_web_search = BraveSearch.from_search_kwargs({"max_results": 5})
 # )
 
 webpage_extraction_tool = TavilyExtract(
-    extract_depth="advanced",
+    extract_depth="basic",
     include_images=True
 )