Update agent.py

agent.py

@@ -1,243 +1,147 @@
-from zoneinfo import ZoneInfo
-from google.adk.agents import Agent,BaseAgent,LlmAgent
-from google.adk.tools import google_search
-from google.adk.runners import Runner
-from google.adk.sessions import InMemorySessionService
-from google.genai import types
-import google.genai.types as types
-import requests
-from google.adk.events import Event, EventActions
-from google.adk.agents.invocation_context import InvocationContext
-from typing import AsyncGenerator
-from google.genai import types as genai_types 
-from google.adk.tools import ToolContext, FunctionTool
-import logging
-#from google.adk.tools import built_in_code_execution
-from google.adk.tools import agent_tool
-
-logging.basicConfig(level=logging.ERROR)
-#from google.adk.tools import agent_tool
-url = 'https://agents-course-unit4-scoring.hf.space/questions'
-headers = {'accept': 'application/json'}
-response = requests.get(url, headers=headers)
-
-# class responses_api(BaseAgent):
-#     async def _run_async_impl(self, ctx: InvocationContext)-> AsyncGenerator[Event, None]:
-#         # This method is called when the agent is run
-#         # You can implement your logic here
-#         # For example, you can call an external API or perform some calculations
-#         # and return the result
-#         url = 'https://agents-course-unit4-scoring.hf.space/questions'
-#         headers = {'accept': 'application/json'}
-#         response = requests.get(url, headers=headers)
-#         for i in response.json():
-#             if i['file_name'] != '':
-#                 url_file = f"https://agents-course-unit4-scoring.hf.space/files/{i['task_id']}"
-#                 question = i['question']
-#                 prompt = f"{question} and the file is {url_file}, give the final answer only"
-#             else:
-#                 question = i['question']
-#                 prompt = f"{question} give the final answer only"
-#             existing_responses = ctx.session.state.get("user:responses", [])
-#             existing_responses.append(prompt)
-#             ctx.session_state["user:responses"] = existing_responses
-
-#             # Optionally, yield a single event to indicate completion or provide some output
-#             yield Event(author=self.name, content=types.Content(parts=[types.Part(text=f"Fetched {len(questions_data)} questions."))])
-
-
-
-def answer_questions():
-    """Fetch questions from the GAIA API and return them in a structured format"""
-    url = 'https://agents-course-unit4-scoring.hf.space/questions'
-    headers = {'accept': 'application/json'}
-    response = requests.get(url, headers=headers)
-
-    if response.status_code != 200:
-        return f"Error fetching questions: {response.status_code}"
-
-    questions_data = response.json()
-    return questions_data
-#responses_api = responses_api(name= 'responses_api_1')
-from typing import Dict, Any
-def submit_questions(answers: list[str]) -> Dict[str, Any]:
-    url = 'https://agents-course-unit4-scoring.hf.space/submit'
-    payload = {
-    "username": "ashishja",
-    "agent_code": "https://huggingface.co/spaces/ashishja/Agents_Course_Final_Assignment_Ashish/tree/main",
-    "answers": answers}
-    headers = {'accept': 'application/json', "Content-Type": "application/json"}
-    response = requests.post(url, headers=headers, json =payload)
-    import json
-    print(json.dumps(payload, indent=2))
-    if response.status_code == 200:
-        return response.json()
-    else:
-        response.raise_for_status()
-
-
-
-
-responses_api = FunctionTool(func= answer_questions)
-submit_api = FunctionTool(func=submit_questions)
-
-# class QuestionAnswerer(LlmAgent):
-#     async def _run_async_impl(self, ctx: InvocationContext) -> AsyncGenerator[Event, None]:
-#         questions_to_answer = ctx.session_service.get('fetched_questions', [])
-#         for q in questions_to_answer:
-#             answer = await self._llm(messages=[types.ChatMessage(role="user", parts=[types.Part(text=q)])])
-#             yield Event(author=self.name, content=answer.content)
-
-# qa = QuestionAnswerer(name = 'qa_1', model="gemini-2.0-flash", description="Question Answerer")
-
-
-
-
-
-
-
-
-APP_NAME="weather_sentiment_agent"
-USER_ID="user1234"
-SESSION_ID="1234"
-
-
-code_agent = LlmAgent(
-    name='codegaiaAgent',
-    model="gemini-2.5-pro-preview-05-06",
-    description=(
-        "You are a smart agent that can write and execute code and answer any questions provided access the given files and answer"
-    ),
-    instruction = (
-"if the question contains a file with .py ,Get the code file and depending on the question and the file provided, execute the code and provide the final answer. "
-"If the question contains a spreadsheet file like .xlsx and .csv among others, get the file and depending on the question and the file provided, execute the code and provide the final answer. "
-"use code like import pandas as pd , file = pd.read_csv('file.csv') and then use the file to answer the question. "
-"if the question contains a file with .txt ,Get the code file and depending on the question and the file provided, execute the code and provide the final answer. "
-"if the question contains a file with .json ,Get the code file and depending on the question and the file provided, execute the code and provide the final answer. "
-"If you are writing code or if you get a code file, use the code execution tool to run the code and provide the final answer. "
-)
-
-,
-    # tools=[built_in_code_execution], 
-    # Add the responses_api agent as a tool
-    #sub_agents=[responses_api]
-)
-
-
-search_agent = LlmAgent(
-    name='searchgaiaAgent',
-    model="gemini-2.5-pro-preview-05-06",
-    description=(
-        "You are a smart agent that can search the web and answer any questions provided access the given files and answer"
-    ),
-    instruction = (
-    "Get the url associated perform a search and consolidate the information provided and answer the provided question "
-)
-
-,
-    tools=[google_search], 
-    # Add the responses_api agent as a tool
-    #sub_agents=[responses_api]
-)
-
-image_agent = LlmAgent(
-    name='imagegaiaAgent',
-    model="gemini-2.5-pro-preview-05-06",
-    description=(
-        "You are a smart agent that can when given a image file and answer any questions related to it"
-    ),
-    instruction = (
-    "Get the image file from the link associated in the prompt use Gemini to watch the video and answer the provided question ")
-
-,
-   # tools=[google_search], 
-    # Add the responses_api agent as a tool
-    #sub_agents=[responses_api]
-)
-
-
-youtube_agent = LlmAgent(
-    name='youtubegaiaAgent',
-    model="gemini-2.5-pro-preview-05-06",
-    description=(
-        "You are a smart agent that can when given a youtube link watch it and answer any questions related to it"
-    ),
-    instruction = (
-    "Get the youtube link associated use Gemini to watch the video and answer the provided question ")
-
-,
-   # tools=[google_search], 
-    # Add the responses_api agent as a tool
-    #sub_agents=[responses_api]
-)
-
-root_agent = LlmAgent(
-    name='basegaiaAgent',
-    model="gemini-2.5-pro-preview-05-06",
-    description=(
-        "You are a smart agent that can answer any questions provided access the given files and answer"
-    ),
-    instruction = (
-    "You are a helpful agent. When the user asks to get the questions or makes a similar request, "
-    "invoke your tool 'responses_api' to retrieve the questions data. "
-    "The questions data will be a list of dictionaries, each containing 'task_id', 'question', and 'file_name' fields. "
-    "For each question in the data: "
-    "1. If file_name is not empty, the file can be accessed at https://agents-course-unit4-scoring.hf.space/files/[TASK_ID] "
-    "2. Use appropriate sub-agents based on question type (code_agent for coding, search_agent for web search, etc.) "
-    "3. Provide a concise, direct answer for each question "
-    "4. Return a dictionary with keys 'task_id' and 'submitted_answer' for each answer "
-    "5. Collect all dictionaries in a list and pass to 'submit_api' tool to submit the answers. "
-    "Always provide direct, factual answers without prefixes like 'The answer is:' or 'Final answer:'"
-)
-
-,
-    tools=[responses_api,submit_api,agent_tool.AgentTool(agent = code_agent),\
-           agent_tool.AgentTool(agent = search_agent), agent_tool.AgentTool(youtube_agent), agent_tool.AgentTool(image_agent)], 
-    # Add the responses_api agent as a tool
-    #sub_agents=[responses_api]
-)
-
-# root_agent = LlmAgent(
-#     name='gaiaAgent',
-#     model="gemini-2.5-pro-preview-05-06",
-#     description=(
-#         "You are a smart agent that can answer any questions provided access the given files and answer"
-#     ),
-#     instruction = (
-#     "You are a helpful agent. When the user asks to get the questions or makes a similar request, "
-#     "invoke base agent. "
-#     "Once you the answers check if are in correct format. "
-#     #"Collect all such dictionaries in a list (do not include any backslashes), and pass this list to the 'submit_api' tool to submit the answers."
-# )
-
-# ,
-#     #tools=[submit_api], 
-#     # Add the responses_api agent as a tool
-#     sub_agents=[base_agent]
-# )
-
-session_service = InMemorySessionService()
-
-# Create the default session synchronously (create_session is not async)
-try:
-    session = session_service.create_session(
-        app_name=APP_NAME,
-        user_id=USER_ID,
-        session_id=SESSION_ID
+from typing import TypedDict, Annotated, Optional
+from langgraph.graph.message import add_messages
+from langchain_core.messages import AnyMessage, HumanMessage, SystemMessage, ToolMessage
+from langgraph.prebuilt import ToolNode, tools_condition
+from langgraph.graph import START, StateGraph, END 
+from langchain_openai import ChatOpenAI
+from pydantic import SecretStr
+import os
+from dotenv import load_dotenv
+from tools import download_file_from_url, basic_web_search, extract_url_content, wikipedia_reader, transcribe_audio_file, question_youtube_video
+
+# Load environment variables from .env file
+load_dotenv()
+
+OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY", "")
+MAIN_LLM_MODEL = os.getenv("MAIN_LLM_MODEL", "google/gemini-2.0-flash-lite-001")
+
+# Generate the chat interface, including the tools
+if not OPENROUTER_API_KEY:
+    raise ValueError("OPENROUTER_API_KEY is not set. Please ensure it is defined in your .env file or environment variables.")
+
+
+def create_agent_graph():
+    
+    main_llm = ChatOpenAI(
+        model=MAIN_LLM_MODEL, # e.g., "mistralai/mistral-7b-instruct"
+        api_key=SecretStr(OPENROUTER_API_KEY), # Your OpenRouter API key
+        base_url="https://openrouter.ai/api/v1", # Standard OpenRouter API base
+        verbose=True # Optional: for debugging
     )
-    print(f"✅ Default session created: {SESSION_ID}")
-except Exception as e:
-    print(f"⚠️ Error creating default session: {e}")
-    session = None
-
-runner = Runner(agent=root_agent, app_name=APP_NAME, session_service=session_service)
-# # def send_query_to_agent(root_agent, query, session):
-# #     session = session
-# #     content = types.Content(role='user', parts=[types.Part(text=query)])
 
-# # async def main():
-# #     await process_questions_and_answer()
 
-# # if __name__ == "__main__":
-# #     import asyncio
-# #     asyncio.run(main())
+    tools = [download_file_from_url, basic_web_search, extract_url_content, wikipedia_reader, transcribe_audio_file, question_youtube_video] # Ensure these tools are defined
+    chat_with_tools = main_llm.bind_tools(tools)
+
+    class AgentState(TypedDict):
+        messages: Annotated[list[AnyMessage], add_messages]
+        file_url: Optional[str | None]
+        file_ext: Optional[str | None]
+        local_file_path: Optional[str | None]
+        final_answer: Optional[str | None]
+    
+    def assistant(state: AgentState):
+        return {
+            "messages": [chat_with_tools.invoke(state["messages"])],
+            "file_url": state.get("file_url", None),
+            "file_ext": state.get("file_ext", None),
+            "local_file_path": state.get("local_file_path", None),
+            "final_answer": state.get("final_answer", None)
+        }
+    
+    def file_path_updater_node(state: AgentState):
+        download_tool_response = state["messages"][-1].content
+        file_path = download_tool_response.split("Local File Path: ")[-1].strip()
+        return {
+            "local_file_path": file_path
+        }
+    
+    def file_path_condition(state: AgentState) -> str:
+        if state["messages"] and isinstance(state["messages"][-1], ToolMessage):
+            tool_response = state["messages"][-1]
+            if tool_response.name == "download_file_from_url":
+                return "update_file_path"  # Route to file path updater if a file was downloaded
+        return "assistant"  # Otherwise, continue with the assistant node
+
+    def format_final_answer_node(state: AgentState) -> AgentState:
+        """
+        Formats the final answer based on the state.
+        This node is reached when the assistant has completed its task.
+        """
+        final_answer = state["messages"][-1].content if state["messages"] else None
+        if final_answer:
+            state["final_answer"] = final_answer.split("FINAL ANSWER:")[-1].strip() #if FINAL_ANSWER isn't present we grab the whole string
+        return state           
+        
+        
+    # The graph
+    builder = StateGraph(AgentState)
+    
+    builder.add_node("assistant", assistant)
+    builder.add_edge(START, "assistant")
+    builder.add_node("tools", ToolNode(tools))
+    builder.add_node("file_path_updater_node", file_path_updater_node)
+    builder.add_node("format_final_answer_node", format_final_answer_node)
+    
+    builder.add_conditional_edges(
+        "assistant",
+        tools_condition,
+        {
+            "tools": "tools", 
+            "__end__": "format_final_answer_node"  # This is the end node for the assistant
+        }
+    )
+    builder.add_conditional_edges(
+        "tools",
+        file_path_condition,
+        {
+            "update_file_path": "file_path_updater_node",
+            "assistant": "assistant"
+        }
+    )
+    
+    builder.add_edge("file_path_updater_node", "assistant")
+    builder.add_edge("format_final_answer_node", END)
+    graph = builder.compile()
+    return graph
+
+class BasicAgent:
+    """
+    A basic agent that can answer questions and download files.
+    Requires a system message be defined in 'system_prompt.txt'.
+    """
+    def __init__(self, graph=None):
+
+        with open("system_prompt.txt", "r", encoding="utf-8") as f:
+            self.system_message = SystemMessage(content=f.read())
+        
+        if graph is None:
+            self.graph = create_agent_graph()
+        else:
+            self.graph = graph
+
+    def __call__(self, question: str, file_url: Optional[str] = None, file_ext: Optional[str] = None) -> str:
+        """
+        Call the agent with a question and optional file URL and extension.
+        
+        Args:
+            question (str): The user's question.
+            file_url (Optional[str]): The URL of the file to download.
+            file_ext (Optional[str]): The file extension for the downloaded file.
+        
+        Returns:
+            str: The agent's response.
+        """
+        if file_url and file_ext:
+            question += f"\nREFERENCE FILE MUST BE RETRIEVED\nFile URL: {file_url}, File Extension: {file_ext}\nUSE A TOOL TO DOWNLOAD THIS FILE."
+        state = {
+            "messages": [self.system_message, HumanMessage(content=question)],
+            "file_url": file_url,
+            "file_ext": file_ext,
+            "local_file_path": None,
+            "final_answer": None
+        }
+        response = self.graph.invoke(state)
+        for m in response["messages"]:
+            m.pretty_print()
+        return response["final_answer"] if response["final_answer"] else "No final answer generated."