Update app.py

app.py

@@ -3,7 +3,7 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
-from smolagents import CodeAgent, InferenceClientModel, DuckDuckGoSearchTool, Tool
+from smolagents import CodeAgent, InferenceClientModel, DuckDuckGoSearchTool, Tool, tool
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -16,37 +16,81 @@ class BasicAgent:
     def __init__(self):
         print("Initializing Smolagent...")
 
-        def fetch_task_file(task_id, api_url=DEFAULT_API_URL):
-            resp = requests.get(f"{api_url}/files/{task_id}")
+        @tool
+        def fetch_task_file(task_id: str) -> str:
+            """
+            Downloads the file attached to a GAIA task and saves it locally.
+
+            Args:
+                task_id: The task_id of the current question.
+
+            Returns:
+                The local file path where the file was saved.
+            """
+            resp = requests.get(f"{DEFAULT_API_URL}/files/{task_id}")
             resp.raise_for_status()
-            return resp.content  # raw image bytes
+            # Try to infer extension from content-type; default to .bin
+            ext = resp.headers.get("content-type", "").split("/")[-1].split(";")[0]
+            path = f"/tmp/{task_id}.{ext or 'bin'}"
+            with open(path, "wb") as f:
+                f.write(resp.content)
+            return path
+
+        @tool
+        def read_spreadsheet(file_path: str) -> str:
+            """
+            Reads a CSV or Excel file and returns a text summary of its contents.
+
+            Args:
+                file_path: Local path to the spreadsheet file.
+
+            Returns:
+                A string representation of the dataframe.
+            """
+            if file_path.endswith(".csv"):
+                df = pd.read_csv(file_path)
+            else:
+                df = pd.read_excel(file_path)
+            return df.to_string()
 
         image_analysis_tool = Tool.from_space(
-            "nvidia/LocateAnything", 
-            name = "image_analyser", 
-            description = "analyses images to look for objects in them", 
-            trust_remote_code = True 
-        )
-        #you need something to pull the spreadhseetso write a custom function for that 
-        
-        spreadsheet_analyser = Tool.from_space(
-            "Abhi8720/spreadsheet-whisperer", 
-            name="spreadsheet_analyser",
-            description="helps answer questions about spreadsheets", 
-            trust_remote_code=True
+            "nvidia/LocateAnything",
+            name="image_analyser",
+            description="Analyses images to locate and describe objects in them.",
+            trust_remote_code=True,
         )
-        #initialising the agent
+
         self.agent = CodeAgent(
-            tools=[DuckDuckGoSearchTool(), fetch_task_file(), image_analysis_tool()], 
-            model=InferenceClientModel("Qwen/Qwen2.5-VL-72B-Instruct"), 
-            instructions=""""You must answer any random question to the best of your knowledge 
-            as use the tools in the following order: 
-            {DuckDuckGoSearchTool}: this is for whenever you need to use the web to answer questions, only use it when you feel you need to 
-            answer questions that require you to search the web
-            {fetch_task_files}: when you need to look for images and fetch them, only use them when you believe the question involves an image
-            {image_analysis_tool}: use it when you need to locate something in an image or video""", 
-            max_steps=10
+            tools=[
+                DuckDuckGoSearchTool(),
+                fetch_task_file,
+                read_spreadsheet,
+                image_analysis_tool,
+            ],
+            model=InferenceClientModel("Qwen/Qwen2.5-VL-72B-Instruct"),
+            instructions=(
+                "Answer with only the final answer value — no explanation, "
+                "no 'FINAL ANSWER' prefix, no extra formatting. "
+                "If the question references a file, first call fetch_task_file "
+                "with the given task_id to download it, then use the appropriate "
+                "tool (read_spreadsheet for tables, image_analyser for images) "
+                "to inspect it before answering. Use DuckDuckGoSearchTool only "
+                "when the question needs current or external information."
+            ),
+            max_steps=10,
         )
+
+    def __call__(self, question: str, task_id: str = None) -> str:
+        print(f"Agent received question: {question[:50]}...")
+        full_prompt = question
+        if task_id:
+            full_prompt = f"{question}\n\n(task_id for any file: {task_id})"
+        try:
+            answer = self.agent.run(full_prompt)
+            return str(answer).strip()
+        except Exception as e:
+            print(f"Error during agent run: {e}")
+            return "Error"
 #answering questions 
     def __call__(self, question: str) -> str:
         print(f"Agent received question: {question[:50]}...")