Add SerpAPI search functionality, audio transcription, Excel analysis, and OCR tools to agent.py. Update requirements.txt to include new dependencies.

agent.py

@@ -12,9 +12,18 @@ import requests
 from bs4 import BeautifulSoup
 import urllib.parse
 import re
+import pandas as pd
+import pytesseract
+from PIL import Image
+import whisper
+import yt_dlp
+import tempfile
+import subprocess
 
 load_dotenv()
 
+SERPAPI_API_KEY = os.getenv("SERPAPI_API_KEY")
+
 @tool
 def wiki_search(query: str) -> str:
     """Search Wikipedia for information.
@@ -33,37 +42,30 @@ def wiki_search(query: str) -> str:
         return f"Error searching Wikipedia: {str(e)}"
 
 @tool
-def web_search(query: str) -> str:
-    """Search the web using DuckDuckGo.
+def serpapi_search(query: str) -> str:
+    """Search the web using SerpAPI (Google Custom Search).
     
     Args:
         query: The search query."""
     try:
-        encoded_query = urllib.parse.quote(query)
-        url = f"https://html.duckduckgo.com/html/?q={encoded_query}"
-        
-        headers = {
-            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
+        if not SERPAPI_API_KEY:
+            return "SerpAPI key not set"
+        params = {
+            "q": query,
+            "api_key": SERPAPI_API_KEY,
+            "engine": "google",
+            "num": 3,
+            "hl": "en"
         }
-        
-        response = requests.get(url, headers=headers)
+        response = requests.get("https://serpapi.com/search", params=params)
         response.raise_for_status()
-        
-        soup = BeautifulSoup(response.text, 'html.parser')
-        
+        data = response.json()
         results = []
-        for result in soup.find_all('div', class_='result__body'):
-            title = result.find('h2', class_='result__title')
-            snippet = result.find('a', class_='result__snippet')
-            
-            if title and snippet:
-                results.append(f"Title: {title.get_text()}\nSnippet: {snippet.get_text()}")
-            
-            if len(results) >= 3:
-                break
-        
+        for r in data.get("organic_results", [])[:3]:
+            title = r.get("title", "")
+            snippet = r.get("snippet", "")
+            results.append(f"Title: {title}\nSnippet: {snippet}")
         return {"web_results": "\n\n".join(results) if results else "No results found"}
-            
     except Exception as e:
         return f"Error searching web: {str(e)}"
 
@@ -92,87 +94,71 @@ def reverse_text(text: str) -> str:
         text: The text to reverse."""
     return text[::-1]
 
+@tool
+def transcribe_audio(file_path: str) -> str:
+    """Transcribe an audio file using Whisper.
+    
+    Args:
+        file_path: Path to the audio file."""
+    try:
+        model = whisper.load_model("base")
+        result = model.transcribe(file_path)
+        return result["text"]
+    except Exception as e:
+        return f"Error transcribing audio: {str(e)}"
+
+@tool
+def analyze_excel(file_path: str, column: str = None) -> str:
+    """Analyze an Excel file and return the sum of a column or all data.
+    
+    Args:
+        file_path: Path to the Excel file.
+        column: Optional column to sum."""
+    try:
+        df = pd.read_excel(file_path)
+        if column and column in df.columns:
+            return str(df[column].sum())
+        return df.to_csv(index=False)
+    except Exception as e:
+        return f"Error analyzing Excel: {str(e)}"
+
+@tool
+def ocr_image(file_path: str) -> str:
+    """Extract text from an image using OCR.
+    
+    Args:
+        file_path: Path to the image file."""
+    try:
+        img = Image.open(file_path)
+        text = pytesseract.image_to_string(img)
+        return text
+    except Exception as e:
+        return f"Error extracting text from image: {str(e)}"
+
 @tool
 def analyze_youtube_video(video_url: str) -> str:
-    """Analyze a YouTube video for specific information.
+    """Download and transcribe a YouTube video using yt-dlp and Whisper.
     
     Args:
         video_url: The URL of the YouTube video."""
     try:
-        # Extract video ID from URL
-        video_id = re.search(r'(?:v=|\/)([0-9A-Za-z_-]{11}).*', video_url)
-        if not video_id:
-            return "Invalid YouTube URL"
-        
-        video_id = video_id.group(1)
-        
-        # Use YouTube API or web search to get video information
-        encoded_query = urllib.parse.quote(f"youtube video {video_id} transcript description")
-        url = f"https://html.duckduckgo.com/html/?q={encoded_query}"
-        
-        headers = {
-            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
-        }
-        
-        response = requests.get(url, headers=headers)
-        response.raise_for_status()
-        
-        soup = BeautifulSoup(response.text, 'html.parser')
-        
-        results = []
-        for result in soup.find_all('div', class_='result__body'):
-            title = result.find('h2', class_='result__title')
-            snippet = result.find('a', class_='result__snippet')
-            
-            if title and snippet:
-                results.append(f"Title: {title.get_text()}\nSnippet: {snippet.get_text()}")
-            
-            if len(results) >= 3:
-                break
-        
-        return {"video_results": "\n\n".join(results) if results else "No results found"}
-            
+        with tempfile.TemporaryDirectory() as tmpdir:
+            ydl_opts = {
+                'format': 'bestaudio/best',
+                'outtmpl': f'{tmpdir}/%(id)s.%(ext)s',
+                'quiet': True,
+            }
+            with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+                info = ydl.extract_info(video_url, download=True)
+                audio_path = ydl.prepare_filename(info)
+            model = whisper.load_model("base")
+            result = model.transcribe(audio_path)
+            return result["text"][:2000]  # Limite la sortie
     except Exception as e:
         return f"Error analyzing YouTube video: {str(e)}"
 
 # System prompt
-system_prompt = """You are a highly accurate question-answering assistant. Your task is to provide precise, direct answers to questions.
-
-Key Rules:
-1. Answer Format:
-   - For numbers: Provide only the number without units, commas, or formatting
-   - For text: Use minimal words, no articles or abbreviations
-   - For lists: Use comma-separated values without additional formatting
-   - For dates: Use YYYY-MM-DD format unless specified otherwise
-   - For names: Use full names without titles or honorifics
-   - For country codes: Use official IOC codes (3 letters)
-   - For chess moves: Use standard algebraic notation
-   - For currency: Use numbers only, no symbols
-
-2. Answer Guidelines:
-   - Be extremely precise and direct
-   - Do not include any explanatory text
-   - Do not use phrases like "FINAL ANSWER" or any markers
-   - Do not include units unless explicitly requested
-   - Do not use abbreviations unless they are standard (e.g., DNA, RNA)
-   - For multiple choice: Provide only the letter or number of the correct answer
-   - For reversed text: Provide the answer in normal text
-   - For file-based questions: Focus on the specific information requested
-
-3. Error Handling:
-   - If uncertain, provide the most likely answer based on available information
-   - If completely unsure, provide a reasonable default rather than an error message
-   - For file processing errors, indicate the specific issue
-
-4. Special Cases:
-   - For mathematical questions: Provide the exact numerical result
-   - For historical dates: Use the most widely accepted date
-   - For scientific terms: Use the standard scientific notation
-   - For geographical locations: Use official names without abbreviations
-   - For audio/video questions: Focus on the specific detail requested
-   - For reversed text: Use the reverse_text tool to decode
-   - For YouTube videos: Use the analyze_youtube_video tool to get information
-   - For chess positions: Provide the move in standard algebraic notation"""
+system_prompt = """You are a highly accurate question-answering assistant. Your answers must be:\n- Direct, with no extra words or explanations.\n- Formatted exactly as requested (numbers only, comma-separated lists, etc.).\n- If the question involves a file, extract only the requested information.\n- If the question is about a video, audio, or image, use the appropriate tool to extract the answer.\n- If you are unsure, provide the most likely answer in the correct format.\n- Never add units, explanations, or formatting unless explicitly requested.\n"""
 
 # System message
 sys_msg = SystemMessage(content=system_prompt)
@@ -180,9 +166,12 @@ sys_msg = SystemMessage(content=system_prompt)
 # Tools list
 tools = [
     wiki_search,
-    web_search,
+    serpapi_search,
     arxiv_search,
     reverse_text,
+    transcribe_audio,
+    analyze_excel,
+    ocr_image,
     analyze_youtube_video,
 ]
 

requirements.txt

@@ -15,3 +15,7 @@ langgraph==0.4.1
 langgraph-checkpoint==2.0.24
 langgraph-prebuilt==0.1.8
 langgraph-sdk==0.1.63
+whisper
+pytesseract
+pillow
+yt-dlp