Create agent.py

agent.py

@@ -0,0 +1,321 @@
+from smolagents import CodeAgent, ToolCallingAgent, LiteLLMModel, tool, Tool, load_tool, WebSearchTool, DuckDuckGoSearchTool
+import asyncio
+import os
+import re
+import pandas as pd
+from typing import Optional
+from token_bucket import Limiter, MemoryStorage
+import yaml
+from PIL import Image, ImageOps
+import requests
+from io import BytesIO
+from markdownify import markdownify
+import whisper
+import time
+import shutil
+import traceback
+from langchain_community.document_loaders import ArxivLoader
+import logging
+import io
+import base64
+
+logger = logging.getLogger(__name__)
+
+@tool
+def search_arxiv(query: str) -> str:
+    """Search Arxiv for a query and return maximum 3 result.
+    
+    Args:
+        query: The search query.
+     Returns:
+        str: Formatted search results
+    """
+    search_docs = ArxivLoader(query=query, load_max_docs=3).load()
+    formatted_search_docs = "\n\n---\n\n".join(
+        [
+            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
+            for doc in search_docs
+        ])
+    return {"arxiv_results": formatted_search_docs}
+
+class ChessboardToFENOnlineTool(Tool):
+    name = "chessboard_to_fen_online"
+    description = "Converts a chessboard image to FEN using an online API (no local templates needed)."
+    inputs = {
+        'image_path': {
+            'type': 'string', 
+            'description': 'Path to the PNG/JPG image of the chessboard.'
+        }
+    }
+    output_type = "string"
+
+    def forward(self, image_path: str) -> str:
+        try:
+            with open(image_path, "rb") as image_file:
+                encoded_image = base64.b64encode(image_file.read()).decode('utf-8')
+        except FileNotFoundError:
+            return "Error: Image file not found."
+
+        api_url = "https://api.chessvision.ai/v1/recognize"
+        headers = {
+            "Authorization": "Bearer YOUR_API_KEY",  # Replace with actual key
+            "Content-Type": "application/json"
+        }
+        payload = {
+            "image": encoded_image,
+            "format": "fen"
+        }
+
+        try:
+            response = requests.post(api_url, headers=headers, json=payload)
+            if response.status_code == 200:
+                return response.json().get("fen", "Error: FEN not found in response.")
+            else:
+                return f"API Error: {response.status_code} - {response.text}"
+        except Exception as e:
+            return f"API Call Failed: {str(e)}"
+
+class VisitWebpageTool(Tool):
+    name = "visit_webpage"
+    description = "Visits a webpage at the given url and reads its content as a markdown string. Use this to browse webpages."
+    inputs = {'url': {'type': 'string', 'description': 'The url of the webpage to visit.'}}
+    output_type = "string"
+
+    @retry(
+        stop=stop_after_attempt(3),
+        wait=wait_exponential(multiplier=1, min=4, max=10),
+        retry=retry_if_exception(is_429_error)
+    )
+    def forward(self, url: str) -> str:
+        try:
+            response = requests.get(url, timeout=50)
+            response.raise_for_status()
+            markdown_content = markdownify(response.text).strip()
+            markdown_content = re.sub(r"\n{3,}", "\n\n", markdown_content)
+            return markdown_content
+        except requests.exceptions.HTTPError as e:
+            if e.response.status_code == 429:
+                raise  # Retry on 429
+            return f"Error fetching the webpage: {str(e)}"
+        except requests.exceptions.Timeout:
+            return "The request timed out. Please try again later or check the URL."
+        except requests.exceptions.RequestException as e:
+            return f"Error fetching the webpage: {str(e)}"
+        except Exception as e:
+            return f"An unexpected error occurred: {str(e)}"
+
+    def __init__(self, *args, **kwargs):
+        self.is_initialized = False
+
+class SpeechToTextTool(Tool):
+    name = "speech_to_text"
+    description = "Converts an audio file to text using OpenAI Whisper."
+    inputs = {
+        "audio_path": {"type": "string", "description": "Path to audio file (.mp3, .wav)"},
+    }
+    output_type = "string"
+
+    def __init__(self):
+        super().__init__()
+        try:
+            self.model = whisper.load_model("base")
+            logger.info("Whisper model loaded successfully.")
+        except Exception as e:
+            logger.error(f"Failed to load Whisper model: {str(e)}")
+            raise RuntimeError(f"Failed to load Whisper model: {str(e)}")
+
+    def forward(self, audio_path: str) -> str:
+        if not os.path.exists(audio_path):
+            return f"Error: File not found at {audio_path}"
+        try:
+            print(f"Starting transcription for {audio_path}...")
+            result = self.model.transcribe(audio_path)
+            print(f"Transcription completed for {audio_path}.")
+            return result.get("text", "")
+        except Exception as e:
+            return f"Error processing audio file: {str(e)}"
+
+class ExcelReaderTool(Tool):
+    name = "excel_reader"
+    description = "Reads and returns a pandas DataFrame from an Excel file (.xlsx, .xls)."
+    inputs = {
+        "excel_path": {
+            "type": "string",
+            "description": "The path to the Excel file to read",
+        },
+        "sheet_name": {
+            "type": "string",
+            "description": "The name of the sheet to read (optional, defaults to first sheet)",
+            "nullable": True
+        }
+    }
+    output_type = "pandas.DataFrame"
+    
+    def forward(self, excel_path: str, sheet_name: str = None) -> pd.DataFrame:
+        try:
+            if not os.path.exists(excel_path):
+                return f"Error: Excel file not found at {excel_path}"
+            if sheet_name:
+                df = pd.read_excel(excel_path, sheet_name=sheet_name)
+            else:
+                df = pd.read_excel(excel_path)
+            return df
+        except Exception as e:
+            return f"Error reading Excel file: {str(e)}"
+
+class PythonCodeReaderTool(Tool):
+    name = "read_python_code"
+    description = "Reads a Python (.py) file and returns its content as a string."
+    inputs = {
+        "file_path": {"type": "string", "description": "The path to the Python file to read"}
+    }
+    output_type = "string"
+
+    def forward(self, file_path: str) -> str:
+        try:
+            if not os.path.exists(file_path):
+                return f"Error: Python file not found at {file_path}"
+            with open(file_path, "r", encoding="utf-8") as file:
+                content = file.read()
+            return content
+        except Exception as e:
+            return f"Error reading Python file: {str(e)}"
+
+from tenacity import retry, stop_after_attempt, wait_exponential, retry_if_exception_type
+
+class RetryDuckDuckGoSearchTool(DuckDuckGoSearchTool):
+    @retry(
+        stop=stop_after_attempt(3),
+        wait=wait_exponential(multiplier=1, min=4, max=10),
+        retry=retry_if_exception_type(Exception)
+    )
+    def forward(self, query: str) -> str:
+        return super().forward(query)
+
+class MagAgent:
+    def __init__(self, rate_limiter: Optional[Limiter] = None):
+        """Initialize the MagAgent with search tools."""
+        logger.info("Initializing MagAgent")
+        self.rate_limiter = rate_limiter
+
+        print("Initializing MagAgent with search tools...")
+        try:
+            # Verify GEMINI_KEY
+            gemini_key = os.environ.get("GEMINI_KEY")
+            if not gemini_key:
+                raise ValueError("GEMINI_KEY environment variable is not set.")
+
+            model = LiteLLMModel(
+                model_id="gemini/gemini-1.5-flash",
+                api_key=gemini_key,
+                max_tokens=8192
+            )
+
+            self.imports = [
+                "pandas",
+                "numpy",
+                "os",
+                "requests",
+                "tempfile",
+                "datetime",
+                "json",
+                "time",
+                "re",
+                "openpyxl",
+                "pathlib",
+                "sys",
+                "bs4",
+                "arxiv",
+                "whisper",
+                "io",
+                "base64"
+            ]
+
+            self.tools = [
+                SpeechToTextTool(),
+                ExcelReaderTool(),
+                PythonCodeReaderTool(),
+                ChessboardToFENOnlineTool(),
+                search_arxiv,
+            ]
+
+            self.prompt_template = (
+                """
+                You are an advanced AI assistant specialized in solving complex, real-world tasks, requiring multi-step reasoning, factual accuracy, and use of external tools.
+                Follow these principles:
+                - Reason step-by-step. Think through the solution logically and plan your actions carefully before answering.
+                - Validate information. Always verify facts when possible instead of guessing.
+                - When processing external data (e.g., YouTube transcripts, web searches), expect potential issues like missing punctuation, inconsistent formatting, or conversational text.
+                - When asked to process Excel files, use the `excel_reader` tool, which returns a pandas DataFrame.
+                - When calculating sales, make sure you multiply volume on price per each product or category.
+                - When asked to transcript YouTube video, try searching it in www.youtubetotranscript.com.
+                - If the input is ambiguous, prioritize extracting key information relevant to the question.
+                - Use code if needed. For calculations, parsing, or transformations, generate Python code and execute it. Be cautious, as some questions contain time-consuming tasks, so analyze the question and choose the most efficient solution.
+                - Be precise and concise. The final answer must strictly match the required format with no extra commentary.
+                - Use tools intelligently. If a question involves external information, structured data, images, or audio, call the appropriate tool to retrieve or process it.
+                - If the question includes direct speech or quoted text (e.g., "Isn't that hot?"), treat it as a precise query and preserve the quoted structure in your response, including quotation marks for direct quotes (e.g., final_answer('"Extremely."')).
+                - If asked about the name of a place or city, use the full complete name without abbreviations (e.g., use Saint Petersburg instead of St.Petersburg).
+                - If asked to look up page numbers, make sure you don't mix them with problem or exercise numbers.
+                - If you cannot retrieve or process data (e.g., due to blocked requests), retry after 15 seconds delay, try another tool (try wikipedia_search, then web_search, then search_arxiv). Otherwise, return a clear error message: "Unable to retrieve data. Search has failed."
+                - Use `final_answer` to give the final answer.
+                
+                QUESTION: {question}
+                
+                {file_section}
+                
+                ANSWER:
+                """
+            )
+
+            web_agent = ToolCallingAgent(
+                tools=[
+                    WebSearchTool(),
+                    VisitWebpageTool(),
+                    search_arxiv,
+                ],
+                model=model,
+                max_steps=15,
+                name="web_search_agent",
+                description="Runs web searches for you.",
+            )        
+            
+            self.agent = CodeAgent(
+                model=model,
+                managed_agents=[web_agent],
+                tools=self.tools,
+                add_base_tools=True,
+                additional_authorized_imports=self.imports,
+                verbosity_level=2,
+                max_steps=10
+            )
+            print("MagAgent initialized.")
+        except Exception as e:
+            logger.error(f"Failed to initialize MagAgent: {str(e)}\n{traceback.format_exc()}")
+            raise
+
+    async def __call__(self, question: str, file_path: Optional[str] = None) -> str:
+        """Process a question asynchronously using the MagAgent."""
+        print(f"MagAgent received question (first 50 chars): {question[:50]}... File path: {file_path}")
+        try:
+            if self.rate_limiter:
+                while not self.rate_limiter.consume(1):
+                    print(f"Rate limit reached. Waiting...")
+                    await asyncio.sleep(4)
+            file_section = f"FILE: {file_path}" if file_path else ""
+            task = self.prompt_template.format(
+                question=question,
+                file_section=file_section
+            )
+            print(f"Calling agent.run...")
+            response = await asyncio.to_thread(self.agent.run, task=task)
+            print(f"Agent.run completed.")
+            response = str(response)
+            if not response:
+                print(f"No answer found.")
+                response = "No answer found."
+            print(f"MagAgent response: {response[:50]}...")
+            return response
+        except Exception as e:
+            error_msg = f"Error processing question: {str(e)}. Check API key or network connectivity."
+            print(error_msg)
+            return error_msg