Spaces:
Sleeping
Sleeping
Fix import error by adding gaia_agent.py module
Browse files- gaia_agent.py +251 -0
gaia_agent.py
ADDED
|
@@ -0,0 +1,251 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
GAIA Agent Implementation
|
| 3 |
+
|
| 4 |
+
This module provides the main GAIA Agent class that interfaces with the app.py file.
|
| 5 |
+
It wraps the core agent functionality in a simpler interface.
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
import logging
|
| 9 |
+
import time
|
| 10 |
+
import os
|
| 11 |
+
import json
|
| 12 |
+
import re
|
| 13 |
+
import traceback
|
| 14 |
+
import hashlib
|
| 15 |
+
from typing import Dict, Any, List, Optional, Union, Tuple, Literal, TypedDict
|
| 16 |
+
|
| 17 |
+
# Import components from agent.py
|
| 18 |
+
from agent import (
|
| 19 |
+
SimpleMemory, WebSearchTool, ContentExtractor, AgentState,
|
| 20 |
+
analyze_question, create_plan, execute_tool, formulate_answer,
|
| 21 |
+
create_fallback_plan
|
| 22 |
+
)
|
| 23 |
+
|
| 24 |
+
# LangChain & LangGraph imports
|
| 25 |
+
from langchain_core.prompts import ChatPromptTemplate
|
| 26 |
+
from langchain_core.output_parsers import StrOutputParser
|
| 27 |
+
from langgraph.graph import StateGraph, END
|
| 28 |
+
from langchain_openai import ChatOpenAI
|
| 29 |
+
|
| 30 |
+
# Configure logging
|
| 31 |
+
logger = logging.getLogger("gaia_agent")
|
| 32 |
+
|
| 33 |
+
# Environment variables
|
| 34 |
+
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
|
| 35 |
+
SERPER_API_KEY = os.getenv("SERPER_API_KEY", "")
|
| 36 |
+
|
| 37 |
+
class GAIAAgent:
|
| 38 |
+
"""
|
| 39 |
+
GAIA (Grounded AI Assistant) agent with web search and content extraction capabilities.
|
| 40 |
+
This class provides a simplified interface for the app.py file to interact with.
|
| 41 |
+
"""
|
| 42 |
+
|
| 43 |
+
def __init__(self):
|
| 44 |
+
"""Initialize the GAIA agent with simplified configuration"""
|
| 45 |
+
self.memory = SimpleMemory()
|
| 46 |
+
logger.info("GAIA Agent initialized")
|
| 47 |
+
|
| 48 |
+
def __call__(self, question: str) -> str:
|
| 49 |
+
"""
|
| 50 |
+
Process a question and generate an answer.
|
| 51 |
+
Compatible with the interface expected by app.py.
|
| 52 |
+
|
| 53 |
+
Args:
|
| 54 |
+
question (str): The question to process
|
| 55 |
+
|
| 56 |
+
Returns:
|
| 57 |
+
str: The answer to the question
|
| 58 |
+
"""
|
| 59 |
+
return self.process_question(question)
|
| 60 |
+
|
| 61 |
+
def process_question(self, question: str) -> str:
|
| 62 |
+
"""
|
| 63 |
+
Process a question and generate an answer.
|
| 64 |
+
|
| 65 |
+
Args:
|
| 66 |
+
question (str): The question to process
|
| 67 |
+
|
| 68 |
+
Returns:
|
| 69 |
+
str: The answer to the question
|
| 70 |
+
"""
|
| 71 |
+
# Generate a cache key for this question
|
| 72 |
+
cache_key = f"question_{hashlib.md5(question.encode()).hexdigest()}"
|
| 73 |
+
|
| 74 |
+
# Check if we have a cached result
|
| 75 |
+
cached_answer = self.memory.get_cached_result(cache_key)
|
| 76 |
+
if cached_answer:
|
| 77 |
+
logger.info(f"Using cached answer for question: {question[:50]}...")
|
| 78 |
+
return cached_answer
|
| 79 |
+
|
| 80 |
+
try:
|
| 81 |
+
# Initialize LLM with error handling
|
| 82 |
+
if OPENAI_API_KEY:
|
| 83 |
+
llm = ChatOpenAI(
|
| 84 |
+
temperature=0,
|
| 85 |
+
model="gpt-3.5-turbo",
|
| 86 |
+
api_key=OPENAI_API_KEY
|
| 87 |
+
)
|
| 88 |
+
|
| 89 |
+
# Process with LangGraph
|
| 90 |
+
answer = self._process_with_langgraph(question, llm)
|
| 91 |
+
else:
|
| 92 |
+
# Fallback to simple pattern-matching responses when LLM is not available
|
| 93 |
+
logger.warning("Using fallback mode (no OpenAI API key provided)")
|
| 94 |
+
answer = self._fallback_processing(question)
|
| 95 |
+
|
| 96 |
+
# Cache the result
|
| 97 |
+
self.memory.cache_result(cache_key, answer)
|
| 98 |
+
|
| 99 |
+
return answer
|
| 100 |
+
|
| 101 |
+
except Exception as e:
|
| 102 |
+
logger.error(f"Error processing question: {str(e)}")
|
| 103 |
+
logger.error(traceback.format_exc())
|
| 104 |
+
|
| 105 |
+
return f"I apologize, but I encountered an error while processing your question: {str(e)}"
|
| 106 |
+
|
| 107 |
+
def _process_with_langgraph(self, question: str, llm) -> str:
|
| 108 |
+
"""Process question using LangGraph workflow"""
|
| 109 |
+
try:
|
| 110 |
+
# Define state transitions
|
| 111 |
+
def should_continue(state: AgentState) -> Literal["continue", "complete"]:
|
| 112 |
+
"""Determine if the agent should continue or is finished"""
|
| 113 |
+
current_step = state.get("current_step", 0)
|
| 114 |
+
plan = state.get("plan", [])
|
| 115 |
+
|
| 116 |
+
# Check if we still have steps to execute
|
| 117 |
+
if current_step is None or plan is None:
|
| 118 |
+
return "complete"
|
| 119 |
+
|
| 120 |
+
if current_step < len(plan):
|
| 121 |
+
return "continue"
|
| 122 |
+
else:
|
| 123 |
+
return "complete"
|
| 124 |
+
|
| 125 |
+
# Create the graph
|
| 126 |
+
workflow = StateGraph(AgentState)
|
| 127 |
+
|
| 128 |
+
# Add nodes
|
| 129 |
+
workflow.add_node("analyze", analyze_question)
|
| 130 |
+
workflow.add_node("create_plan", create_plan)
|
| 131 |
+
workflow.add_node("execute_tool", execute_tool)
|
| 132 |
+
workflow.add_node("formulate_answer", formulate_answer)
|
| 133 |
+
|
| 134 |
+
# Add edges
|
| 135 |
+
workflow.add_edge("analyze", "create_plan")
|
| 136 |
+
workflow.add_edge("create_plan", "execute_tool")
|
| 137 |
+
workflow.add_edge("execute_tool", should_continue)
|
| 138 |
+
workflow.add_conditional_edges(
|
| 139 |
+
"execute_tool",
|
| 140 |
+
should_continue,
|
| 141 |
+
{
|
| 142 |
+
"continue": "execute_tool",
|
| 143 |
+
"complete": "formulate_answer"
|
| 144 |
+
}
|
| 145 |
+
)
|
| 146 |
+
workflow.add_edge("formulate_answer", END)
|
| 147 |
+
|
| 148 |
+
# Set entry point
|
| 149 |
+
workflow.set_entry_point("analyze")
|
| 150 |
+
|
| 151 |
+
# Compile the graph
|
| 152 |
+
app = workflow.compile()
|
| 153 |
+
|
| 154 |
+
# Run the graph
|
| 155 |
+
state = {
|
| 156 |
+
"question": question,
|
| 157 |
+
"tool_results": []
|
| 158 |
+
}
|
| 159 |
+
|
| 160 |
+
result = app.invoke({
|
| 161 |
+
**state,
|
| 162 |
+
"llm": llm
|
| 163 |
+
})
|
| 164 |
+
|
| 165 |
+
# Return the answer or an error message
|
| 166 |
+
if "answer" in result and result["answer"]:
|
| 167 |
+
return result["answer"]
|
| 168 |
+
elif "error" in result and result["error"]:
|
| 169 |
+
return f"I encountered an error: {result['error']}"
|
| 170 |
+
else:
|
| 171 |
+
return "I was unable to generate an answer based on the available information."
|
| 172 |
+
|
| 173 |
+
except Exception as e:
|
| 174 |
+
logger.error(f"Error in LangGraph processing: {str(e)}")
|
| 175 |
+
logger.error(traceback.format_exc())
|
| 176 |
+
return f"I encountered an error while processing your question: {str(e)}"
|
| 177 |
+
|
| 178 |
+
def _fallback_processing(self, question: str) -> str:
|
| 179 |
+
"""Simple fallback implementation when LLM is not available"""
|
| 180 |
+
try:
|
| 181 |
+
# Simple pattern matching for some question types
|
| 182 |
+
if "how" in question.lower():
|
| 183 |
+
answer = f"To address '{question.strip('?')}', I would recommend following these steps: 1) Understand the core concepts, 2) Apply a structured approach, 3) Evaluate results, and 4) Refine as needed. Without being able to access external knowledge at the moment, this is a general framework for addressing how-to questions."
|
| 184 |
+
elif "what" in question.lower():
|
| 185 |
+
answer = f"Regarding '{question.strip('?')}', this typically involves understanding several key factors. While I don't have access to external knowledge at the moment, this type of question usually requires defining terms, establishing context, and examining relevant concepts."
|
| 186 |
+
elif "why" in question.lower():
|
| 187 |
+
answer = f"The question '{question.strip('?')}' relates to causality and explanation. Such questions typically involve understanding underlying mechanisms, historical context, and logical relationships between factors."
|
| 188 |
+
else:
|
| 189 |
+
# Try a web search if available
|
| 190 |
+
try:
|
| 191 |
+
search_tool = WebSearchTool()
|
| 192 |
+
search_results = search_tool.search(question)
|
| 193 |
+
|
| 194 |
+
if search_results and search_results[0].get("snippet"):
|
| 195 |
+
snippet = search_results[0]["snippet"]
|
| 196 |
+
answer = f"Based on available information: {snippet}\n\nPlease note that without access to a language model, I can only provide this basic search result."
|
| 197 |
+
else:
|
| 198 |
+
answer = f"I'm sorry, but I cannot provide a comprehensive answer to '{question}' at this moment due to limited access to external knowledge and language model capabilities."
|
| 199 |
+
except Exception as search_err:
|
| 200 |
+
logger.error(f"Error in fallback search: {str(search_err)}")
|
| 201 |
+
answer = f"I'm sorry, but I cannot provide a comprehensive answer to '{question}' at this moment due to limited access to external knowledge and language model capabilities."
|
| 202 |
+
|
| 203 |
+
return answer
|
| 204 |
+
except Exception as e:
|
| 205 |
+
logger.error(f"Error in fallback processing: {str(e)}")
|
| 206 |
+
return f"I apologize, but I'm currently unable to process your question due to system limitations."
|
| 207 |
+
|
| 208 |
+
def query(self, question: str) -> Dict[str, Any]:
|
| 209 |
+
"""
|
| 210 |
+
Query the agent with a question to get an answer with metadata.
|
| 211 |
+
|
| 212 |
+
Args:
|
| 213 |
+
question (str): The question to answer
|
| 214 |
+
|
| 215 |
+
Returns:
|
| 216 |
+
Dict[str, Any]: Dictionary containing the answer and metadata
|
| 217 |
+
"""
|
| 218 |
+
try:
|
| 219 |
+
# Track timing
|
| 220 |
+
start_time = time.time()
|
| 221 |
+
|
| 222 |
+
# Process the question
|
| 223 |
+
answer = self.process_question(question)
|
| 224 |
+
|
| 225 |
+
# Calculate processing time
|
| 226 |
+
processing_time = time.time() - start_time
|
| 227 |
+
|
| 228 |
+
# Return result with metadata
|
| 229 |
+
return {
|
| 230 |
+
"question": question,
|
| 231 |
+
"answer": answer,
|
| 232 |
+
"processing_time": processing_time,
|
| 233 |
+
"timestamp": time.time(),
|
| 234 |
+
"status": "success"
|
| 235 |
+
}
|
| 236 |
+
except Exception as e:
|
| 237 |
+
logger.error(f"Error in query: {str(e)}")
|
| 238 |
+
|
| 239 |
+
return {
|
| 240 |
+
"question": question,
|
| 241 |
+
"answer": f"Error processing query: {str(e)}",
|
| 242 |
+
"processing_time": time.time() - start_time,
|
| 243 |
+
"timestamp": time.time(),
|
| 244 |
+
"status": "error",
|
| 245 |
+
"error": str(e)
|
| 246 |
+
}
|
| 247 |
+
|
| 248 |
+
def clear_memory(self):
|
| 249 |
+
"""Clear the agent's memory"""
|
| 250 |
+
self.memory.clear()
|
| 251 |
+
logger.info("Agent memory cleared")
|