JoachimVC's picture
Implement full GAIA agent solution with formatter and multimodal processing
460ec88
"""
GAIA Agent Enhanced Implementation
This module provides an enhanced implementation of the GAIA agent
that uses specialized components instead of hardcoded responses.
"""
import os
import re
import logging
import time
from typing import Dict, Any, List, Optional, Union, Callable
import traceback
import sys
import json
# Import answer formatter
from src.gaia.agent.answer_formatter import format_answer_by_type
# Import LangGraph components
try:
from langgraph.graph import END, StateGraph
from langgraph.prebuilt import ToolNode
LANGGRAPH_AVAILABLE = True
except ImportError:
LANGGRAPH_AVAILABLE = False
# Set up logging
logger = logging.getLogger("gaia_agent")
# Import specialized components
from src.gaia.agent.components import TextAnalyzer, VideoAnalyzer, SearchManager, MemoryManager
from src.gaia.agent.tool_registry import get_tools, create_tools_registry
from src.gaia.agent.config import VERBOSE, DEFAULT_CHECKPOINT_PATH
class GAIAAgent:
"""
Enhanced GAIA Agent implementation.
This agent uses specialized components to handle different types of questions
without hardcoded responses.
"""
def __init__(self, config: Optional[Dict[str, Any]] = None):
"""
Initialize the GAIA Agent with configuration.
Args:
config: Configuration dictionary
"""
self.config = config or {}
self.verbose = self.config.get("verbose", VERBOSE)
# Initialize components
self._initialize_components()
# Initialize state
self.state = {
"initialized": True,
"last_question": None,
"last_answer": None,
"last_execution_time": None
}
# Initialize LangGraph if available
if LANGGRAPH_AVAILABLE:
self.graph = self._build_langgraph()
logger.info("LangGraph workflow initialized")
else:
self.graph = None
logger.warning("LangGraph not available, using fallback processing")
# Tool registry
self.tools_registry = create_tools_registry()
logger.info("GAIA Agent initialized successfully")
def _initialize_components(self):
"""Initialize specialized components."""
logger.info("Initializing components")
try:
# Text Analysis component
self.text_analyzer = TextAnalyzer()
# Video Analysis component
self.video_analyzer = VideoAnalyzer()
# Search component
self.search_manager = SearchManager(self.config.get("search", {}))
# Memory component
self.memory_manager = MemoryManager(self.config.get("memory", {
"use_supabase": bool(os.getenv("SUPABASE_URL", "")),
"cache_enabled": True
}))
logger.info("All components initialized successfully")
except Exception as e:
logger.error(f"Error initializing components: {str(e)}")
logger.debug(traceback.format_exc())
raise RuntimeError(f"Failed to initialize GAIA Agent components: {str(e)}")
def _build_langgraph(self) -> Optional[StateGraph]:
"""
Build and return the LangGraph workflow.
Returns:
StateGraph or None if LangGraph is unavailable
"""
if not LANGGRAPH_AVAILABLE:
return None
try:
from src.gaia.agent.graph import build_agent_graph
return build_agent_graph()
except Exception as e:
logger.error(f"Error building LangGraph: {str(e)}")
logger.debug(traceback.format_exc())
return None
def _detect_question_type(self, question: str) -> str:
"""
Detect the type of question to determine appropriate handling.
Args:
question: The question to analyze
Returns:
str: Question type identifier
"""
question_lower = question.lower()
# Check for reversed text
if self.text_analyzer.is_reversed_text(question):
return "reversed_text"
# Check for scrambled words (all caps is a clue in assessment)
if re.search(r'\b[A-Z]{4,}\b', question):
return "unscramble_word"
# Check for YouTube video questions
if "youtube.com/watch" in question_lower or "youtu.be/" in question_lower:
return "youtube_video"
# Check for image analysis questions
if "image" in question_lower and ("analyze" in question_lower or "what" in question_lower or "describe" in question_lower):
return "image_analysis"
# Check for audio analysis questions
if ".mp3" in question_lower or "audio" in question_lower or "recording" in question_lower:
return "audio_analysis"
# Check for chess position questions
if "chess" in question_lower and "position" in question_lower:
return "chess_analysis"
# Check for mathematical operations
if re.search(r'(\d+\s*[\+\-\*\/]\s*\d+)', question_lower) or "calculate" in question_lower:
return "math_question"
# Default to general knowledge
return "general_knowledge"
def process_question(self, question: str) -> str:
"""
Process a question using appropriate components based on question type.
Args:
question: The question to process
Returns:
str: The generated answer
"""
start_time = time.time()
logger.info(f"Processing question: {question}")
try:
# Check cache first
cached_answer = self.memory_manager.get_cached_answer(question)
if cached_answer:
logger.info("Retrieved answer from cache")
# Update state
self.state["last_question"] = question
self.state["last_answer"] = cached_answer
self.state["last_execution_time"] = time.time() - start_time
return cached_answer
# Detect question type
question_type = self._detect_question_type(question)
logger.info(f"Detected question type: {question_type}")
# Process based on question type
answer = None
# Handle reversed text questions
if question_type == "reversed_text":
logger.info("Processing reversed text question")
text_analysis = self.text_analyzer.process_text_question(question)
if text_analysis.get("answer"):
answer = text_analysis["answer"]
else:
# Fallback to general processing
logger.info("Specialized handling failed, trying general processing")
answer = self._process_with_langgraph(question)
# Handle word unscrambling
elif question_type == "unscramble_word":
logger.info("Processing word unscrambling question")
text_analysis = self.text_analyzer.process_text_question(question)
if text_analysis.get("answer"):
answer = text_analysis["answer"]
else:
# Fallback to general processing
logger.info("Specialized handling failed, trying general processing")
answer = self._process_with_langgraph(question)
# Handle YouTube video questions
elif question_type == "youtube_video":
logger.info("Processing YouTube video question")
# Extract video URL or ID
video_url_match = re.search(r'((?:https?:\/\/)?(?:www\.)?(?:youtube\.com\/watch\?v=|youtu\.be\/)[a-zA-Z0-9_-]+)', question)
if video_url_match:
video_url = video_url_match.group(1)
video_analysis = self.video_analyzer.analyze_video_content(video_url, question)
if video_analysis.get("answer"):
answer = video_analysis["answer"]
else:
# Fallback to general processing
logger.info("Video analysis failed, trying general processing")
answer = self._process_with_langgraph(question)
else:
# No video URL found
logger.warning("No YouTube URL found in question")
answer = "I couldn't find a YouTube video URL in your question. Please provide a valid YouTube link for analysis."
# Handle audio analysis (e.g., MP3 files mentioned in the question)
elif question_type == "audio_analysis":
logger.info("Processing audio analysis question")
# For audio analysis, we currently need to fall back to LangGraph processing
# This could be enhanced with a dedicated AudioAnalyzer component in the future
answer = self._process_with_langgraph(question)
# Handle image analysis (including chess positions)
elif question_type in ["image_analysis", "chess_analysis"]:
logger.info(f"Processing {question_type} question")
# Image and chess analysis are handled by LangGraph and multimodal tools
answer = self._process_with_langgraph(question)
# Handle math questions with direct calculation
elif question_type == "math_question":
logger.info("Processing math question")
# Try to extract and calculate simple expressions
# This is a simplified implementation - complex math would go to LangGraph
expression_match = re.search(r'(\d+)\s*([\+\-\*\/])\s*(\d+)', question)
if expression_match:
try:
num1 = int(expression_match.group(1))
operator = expression_match.group(2)
num2 = int(expression_match.group(3))
result = None
if operator == '+':
result = num1 + num2
elif operator == '-':
result = num1 - num2
elif operator == '*':
result = num1 * num2
elif operator == '/' and num2 != 0:
result = num1 / num2
if result is not None:
answer = f"The result of {num1} {operator} {num2} is {result}."
else:
# Fallback to LangGraph for complex math
answer = self._process_with_langgraph(question)
except Exception:
# Fallback to LangGraph for complex math
answer = self._process_with_langgraph(question)
else:
# Fallback to LangGraph for complex math
answer = self._process_with_langgraph(question)
# Default to general knowledge processing
else:
logger.info("Processing general knowledge question")
answer = self._process_with_langgraph(question)
# If LangGraph processing failed or returned None, use search as fallback
if not answer:
logger.warning("LangGraph processing failed, using search fallback")
search_result = self.search_manager.search(question)
answer = search_result.get("answer", "I couldn't find a specific answer to your question.")
# Cache the question-answer pair
self.memory_manager.cache_question_answer(question, answer)
# Format the answer according to GAIA benchmark requirements
formatted_answer = format_answer_by_type(answer, question)
# Update state
self.state["last_question"] = question
self.state["last_answer"] = formatted_answer
self.state["last_execution_time"] = time.time() - start_time
logger.info(f"Question processed in {time.time() - start_time:.2f} seconds")
logger.debug(f"Original answer: {answer}")
logger.debug(f"Formatted answer: {formatted_answer}")
return formatted_answer
except Exception as e:
logger.error(f"Error processing question: {str(e)}")
logger.debug(traceback.format_exc())
# Provide a graceful error message (without prefixes)
error_msg = f"Error processing the question. Please try rephrasing it."
# Only include technical details in debug/development environments
if self.verbose:
error_msg = f"Error: {str(e)}"
return error_msg
def _process_with_langgraph(self, question: str) -> Optional[str]:
"""
Process a question using the LangGraph workflow.
Args:
question: The question to process
Returns:
str or None: Generated answer or None if processing failed
"""
if not self.graph:
logger.warning("LangGraph not available, using search fallback")
search_result = self.search_manager.search(question)
return search_result.get("answer")
try:
logger.info("Processing with LangGraph workflow")
# Prepare input state for the graph
input_state = {
"question": question,
"tools": get_tools(), # Get the available tools
"thoughts": [],
"messages": [],
"answer": None,
"tool_results": {}
}
# Run the graph
result = self.graph.invoke(input_state)
if result and "answer" in result:
answer = result["answer"]
# Format the answer for LangGraph responses too
formatted_answer = format_answer_by_type(answer, question)
logger.info("Successfully processed with LangGraph")
logger.debug(f"Original LangGraph answer: {answer}")
logger.debug(f"Formatted LangGraph answer: {formatted_answer}")
return formatted_answer
else:
logger.warning("LangGraph processing did not produce an answer")
return None
except Exception as e:
logger.error(f"Error in LangGraph processing: {str(e)}")
logger.debug(traceback.format_exc())
return None
def run(self, input_data: Union[Dict[str, Any], str]) -> str:
"""
Run the agent on the provided input data.
Args:
input_data: Either a dictionary containing the question or the question string directly
Returns:
str: Generated answer
"""
# Handle both string and dictionary inputs
if isinstance(input_data, str):
question = input_data
else:
# Handle dictionary input
question = input_data.get("question", "")
if not question:
return "No question provided. Please provide a question to get a response."
return self.process_question(question)
def get_state(self) -> Dict[str, Any]:
"""
Get the current state of the agent.
Returns:
dict: Current agent state
"""
return self.state.copy()
def reset(self) -> None:
"""Reset the agent state."""
logger.info("Resetting agent state")
# Reset state
self.state = {
"initialized": True,
"last_question": None,
"last_answer": None,
"last_execution_time": None
}
# Clear cache if requested in config
if self.config.get("clear_cache_on_reset", False):
self.memory_manager.clear_cache()