Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,30 +1,31 @@
|
|
| 1 |
"""LangGraph Agent with Gradio Interface"""
|
| 2 |
import os
|
| 3 |
-
import base64
|
| 4 |
-
import gradio as gr
|
| 5 |
-
import requests
|
| 6 |
-
import pandas as pd
|
| 7 |
from dotenv import load_dotenv
|
| 8 |
-
import google.generativeai as genai
|
| 9 |
from langgraph.graph import START, StateGraph, MessagesState
|
| 10 |
-
from langgraph.prebuilt import tools_condition
|
|
|
|
| 11 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
| 12 |
-
from
|
| 13 |
-
from
|
| 14 |
-
from langchain_community.
|
|
|
|
|
|
|
|
|
|
| 15 |
from langchain_core.messages import SystemMessage, HumanMessage
|
| 16 |
from langchain_core.tools import tool
|
| 17 |
from langchain.tools.retriever import create_retriever_tool
|
| 18 |
-
from
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
from typing import
|
|
|
|
| 22 |
|
| 23 |
# Load environment variables
|
| 24 |
load_dotenv()
|
| 25 |
|
| 26 |
# Setup API keys
|
| 27 |
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
|
|
|
|
| 28 |
OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
|
| 29 |
|
| 30 |
if GOOGLE_API_KEY:
|
|
@@ -32,328 +33,509 @@ if GOOGLE_API_KEY:
|
|
| 32 |
else:
|
| 33 |
print("Warning: GOOGLE_API_KEY not found")
|
| 34 |
|
| 35 |
-
|
| 36 |
@tool
|
| 37 |
def multiply(a: int, b: int) -> int:
|
| 38 |
-
"""Multiply two numbers.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
return a * b
|
| 40 |
|
| 41 |
@tool
|
| 42 |
def add(a: int, b: int) -> int:
|
| 43 |
-
"""Add two numbers.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
return a + b
|
| 45 |
|
| 46 |
-
@tool
|
| 47 |
-
def modulus(a: int, b: int) -> int:
|
| 48 |
-
"""Get the modulus of two numbers."""
|
| 49 |
-
return a % b
|
| 50 |
-
|
| 51 |
@tool
|
| 52 |
def subtract(a: int, b: int) -> int:
|
| 53 |
-
"""Subtract two numbers.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
return a - b
|
| 55 |
|
| 56 |
@tool
|
| 57 |
def divide(a: int, b: int) -> int:
|
| 58 |
-
"""Divide two numbers.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
if b == 0:
|
| 60 |
raise ValueError("Cannot divide by zero.")
|
| 61 |
return a / b
|
| 62 |
|
| 63 |
@tool
|
| 64 |
-
def
|
| 65 |
-
"""
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
except Exception as e:
|
| 73 |
-
return {"arvix_results": f"Error: {str(e)}"}
|
| 74 |
-
|
| 75 |
-
@tool
|
| 76 |
-
def execute_python(code: str) -> str:
|
| 77 |
-
"""Execute Python code securely and return results. Handles calculations and data analysis."""
|
| 78 |
-
try:
|
| 79 |
-
parsed = ast.parse(code)
|
| 80 |
-
if any(isinstance(node, (ast.Import, ast.ImportFrom)) for node in parsed.body):
|
| 81 |
-
return "Cannot import modules for security reasons"
|
| 82 |
-
|
| 83 |
-
restricted_builtins = {'open', 'eval', 'exec', '__import__'}
|
| 84 |
-
for node in ast.walk(parsed):
|
| 85 |
-
if isinstance(node, ast.Name) and node.id in restricted_builtins:
|
| 86 |
-
return "Restricted function used"
|
| 87 |
-
|
| 88 |
-
return str(eval(code, {"__builtins__": {}}, {}))
|
| 89 |
-
except Exception as e:
|
| 90 |
-
return f"Execution error: {str(e)}"
|
| 91 |
-
|
| 92 |
-
@tool
|
| 93 |
-
def process_file(file_path: str) -> Dict[str, Any]:
|
| 94 |
-
"""Process uploaded files (Excel, CSV, TXT) and extract structured data."""
|
| 95 |
-
try:
|
| 96 |
-
if file_path.endswith(('.xlsx', '.xls')):
|
| 97 |
-
df = pd.read_excel(file_path)
|
| 98 |
-
return {"data": df.head(10).to_dict(), "summary": df.describe().to_dict()}
|
| 99 |
-
elif file_path.endswith('.csv'):
|
| 100 |
-
df = pd.read_csv(file_path)
|
| 101 |
-
return {"data": df.head(10).to_dict(), "summary": df.describe().to_dict()}
|
| 102 |
-
elif file_path.endswith('.txt'):
|
| 103 |
-
with open(file_path, 'r') as f:
|
| 104 |
-
content = f.read(2000)
|
| 105 |
-
return {"content": content}
|
| 106 |
-
return {"error": "Unsupported file format"}
|
| 107 |
-
except Exception as e:
|
| 108 |
-
return {"error": str(e)}
|
| 109 |
|
| 110 |
@tool
|
| 111 |
-
def
|
| 112 |
-
"""
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
f"
|
| 120 |
-
for doc in
|
| 121 |
])
|
| 122 |
-
|
| 123 |
-
return f"Wikipedia Error: {str(e)}"
|
| 124 |
|
| 125 |
@tool
|
| 126 |
-
def
|
| 127 |
-
"""
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
|
|
|
|
|
|
| 134 |
])
|
| 135 |
-
|
| 136 |
-
return f"Media Search Error: {str(e)}"
|
| 137 |
|
| 138 |
@tool
|
| 139 |
-
def
|
| 140 |
-
"""Search
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
|
| 148 |
|
| 149 |
-
@tool
|
| 150 |
-
def web_search(query: str) -> str:
|
| 151 |
-
"""Search DuckDuckGo for a query and return maximum 3 results."""
|
| 152 |
-
try:
|
| 153 |
-
search = DuckDuckGoSearchResults(max_results=3)
|
| 154 |
-
search_docs = search.invoke(query)
|
| 155 |
-
formatted_search_docs = "\n\n---\n\n".join(
|
| 156 |
-
[f'<Document source="{doc["link"]}"/>\n{doc["snippet"]}\n</Document>'
|
| 157 |
-
for doc in search_docs])
|
| 158 |
-
return {"web_results": formatted_search_docs}
|
| 159 |
-
except Exception as e:
|
| 160 |
-
return {"web_results": f"Error: {str(e)}"}
|
| 161 |
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
if not text or len(text) < 100:
|
| 166 |
-
return "Text too short to summarize."
|
| 167 |
-
return f"Summary of the provided text with key points."
|
| 168 |
|
| 169 |
-
|
| 170 |
-
def parse_query(query: str) -> dict:
|
| 171 |
-
"""Parse a complex query into its key components for better search."""
|
| 172 |
-
parts = query.split()
|
| 173 |
-
return {
|
| 174 |
-
"main_topic": parts[0] if parts else "",
|
| 175 |
-
"subtopics": parts[1:3] if len(parts) > 1 else [],
|
| 176 |
-
"context": " ".join(parts[3:]) if len(parts) > 3 else ""
|
| 177 |
-
}
|
| 178 |
-
|
| 179 |
-
# System Prompt Setup
|
| 180 |
-
system_prompt = """You are a POWERFUL assistant REQUIRED to answer ALL questions using available tools.
|
| 181 |
-
STRICT RULES:
|
| 182 |
-
1. NEVER say you can't answer - ALWAYS use tools
|
| 183 |
-
2. Combine information from multiple tools when needed
|
| 184 |
-
3. For calculations, use execute_python
|
| 185 |
-
4. For files, use process_file
|
| 186 |
-
5. For media/celebrities, use media_search
|
| 187 |
-
6. For academic content, use academic_search
|
| 188 |
-
7. ALWAYS format final answer as: FINAL ANSWER: [your answer]
|
| 189 |
-
|
| 190 |
-
AVAILABLE TOOLS:
|
| 191 |
-
- execute_python: Math/code execution
|
| 192 |
-
- process_file: Analyze uploaded files
|
| 193 |
-
- enhanced_wiki_search: Full Wikipedia access
|
| 194 |
-
- media_search: Videos/images/celebrities
|
| 195 |
-
- academic_search: Textbooks/papers
|
| 196 |
-
- web_search: General web search
|
| 197 |
-
- vector_store: Previous knowledge
|
| 198 |
-
|
| 199 |
-
YOU MUST USE THESE TOOLS TO ANSWER ALL QUESTIONS!"""
|
| 200 |
sys_msg = SystemMessage(content=system_prompt)
|
| 201 |
|
| 202 |
-
#
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 213 |
|
| 214 |
-
# Tool Configuration
|
| 215 |
tools = [
|
| 216 |
-
multiply,
|
| 217 |
-
|
| 218 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 219 |
]
|
| 220 |
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 228 |
)
|
|
|
|
| 229 |
|
| 230 |
-
#
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
-
|
| 251 |
-
|
| 252 |
-
def
|
| 253 |
-
|
| 254 |
-
|
| 255 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 256 |
|
| 257 |
-
|
| 258 |
-
|
| 259 |
-
|
| 260 |
-
|
| 261 |
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
|
| 287 |
-
|
| 288 |
-
|
| 289 |
-
|
| 290 |
-
|
| 291 |
-
|
| 292 |
-
# Graph Builder
|
| 293 |
-
def build_graph():
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
|
| 298 |
|
| 299 |
-
|
| 300 |
|
| 301 |
-
|
| 302 |
-
|
| 303 |
|
| 304 |
-
|
| 305 |
|
| 306 |
-
|
| 307 |
-
|
| 308 |
-
|
| 309 |
-
|
| 310 |
-
|
| 311 |
-
|
| 312 |
|
| 313 |
-
|
| 314 |
-
|
| 315 |
-
|
| 316 |
-
|
| 317 |
|
| 318 |
-
|
| 319 |
-
|
| 320 |
-
|
| 321 |
-
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
|
| 326 |
-
|
| 327 |
-
|
| 328 |
-
|
| 329 |
-
|
| 330 |
-
|
| 331 |
-
|
| 332 |
-
|
| 333 |
-
|
| 334 |
-
|
| 335 |
-
|
| 336 |
-
|
| 337 |
-
|
| 338 |
-
|
| 339 |
-
|
| 340 |
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
|
| 344 |
-
|
| 345 |
|
| 346 |
-
|
| 347 |
-
|
| 348 |
-
|
| 349 |
-
|
| 350 |
|
| 351 |
-
|
| 352 |
-
|
| 353 |
-
|
| 354 |
-
|
| 355 |
|
| 356 |
-
|
| 357 |
|
| 358 |
# Agent Class
|
| 359 |
class BasicAgent:
|
|
|
|
| 1 |
"""LangGraph Agent with Gradio Interface"""
|
| 2 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
from dotenv import load_dotenv
|
|
|
|
| 4 |
from langgraph.graph import START, StateGraph, MessagesState
|
| 5 |
+
from langgraph.prebuilt import tools_condition
|
| 6 |
+
from langgraph.prebuilt import ToolNode
|
| 7 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
| 8 |
+
from langchain_groq import ChatGroq
|
| 9 |
+
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint, HuggingFaceEmbeddings
|
| 10 |
+
from langchain_community.tools.tavily_search import TavilySearchResults
|
| 11 |
+
from langchain_community.document_loaders import WikipediaLoader
|
| 12 |
+
from langchain_community.document_loaders import ArxivLoader
|
| 13 |
+
from langchain_community.vectorstores import SupabaseVectorStore
|
| 14 |
from langchain_core.messages import SystemMessage, HumanMessage
|
| 15 |
from langchain_core.tools import tool
|
| 16 |
from langchain.tools.retriever import create_retriever_tool
|
| 17 |
+
from supabase.client import Client, create_client
|
| 18 |
+
import json
|
| 19 |
+
from typing import Literal
|
| 20 |
+
from typing import TypedDict, Annotated, Sequence
|
| 21 |
+
import operator
|
| 22 |
|
| 23 |
# Load environment variables
|
| 24 |
load_dotenv()
|
| 25 |
|
| 26 |
# Setup API keys
|
| 27 |
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
|
| 28 |
+
HF_API_TOKEN = os.getenv("HF_API_TOKEN")
|
| 29 |
OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
|
| 30 |
|
| 31 |
if GOOGLE_API_KEY:
|
|
|
|
| 33 |
else:
|
| 34 |
print("Warning: GOOGLE_API_KEY not found")
|
| 35 |
|
| 36 |
+
|
| 37 |
@tool
|
| 38 |
def multiply(a: int, b: int) -> int:
|
| 39 |
+
"""Multiply two numbers.
|
| 40 |
+
Args:
|
| 41 |
+
a: first int
|
| 42 |
+
b: second int
|
| 43 |
+
"""
|
| 44 |
return a * b
|
| 45 |
|
| 46 |
@tool
|
| 47 |
def add(a: int, b: int) -> int:
|
| 48 |
+
"""Add two numbers.
|
| 49 |
+
|
| 50 |
+
Args:
|
| 51 |
+
a: first int
|
| 52 |
+
b: second int
|
| 53 |
+
"""
|
| 54 |
return a + b
|
| 55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
@tool
|
| 57 |
def subtract(a: int, b: int) -> int:
|
| 58 |
+
"""Subtract two numbers.
|
| 59 |
+
|
| 60 |
+
Args:
|
| 61 |
+
a: first int
|
| 62 |
+
b: second int
|
| 63 |
+
"""
|
| 64 |
return a - b
|
| 65 |
|
| 66 |
@tool
|
| 67 |
def divide(a: int, b: int) -> int:
|
| 68 |
+
"""Divide two numbers.
|
| 69 |
+
|
| 70 |
+
Args:
|
| 71 |
+
a: first int
|
| 72 |
+
b: second int
|
| 73 |
+
"""
|
| 74 |
if b == 0:
|
| 75 |
raise ValueError("Cannot divide by zero.")
|
| 76 |
return a / b
|
| 77 |
|
| 78 |
@tool
|
| 79 |
+
def modulus(a: int, b: int) -> int:
|
| 80 |
+
"""Get the modulus of two numbers.
|
| 81 |
+
|
| 82 |
+
Args:
|
| 83 |
+
a: first int
|
| 84 |
+
b: second int
|
| 85 |
+
"""
|
| 86 |
+
return a % b
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
|
| 88 |
@tool
|
| 89 |
+
def wiki_search(query: str) -> str:
|
| 90 |
+
"""Search Wikipedia for a query and return maximum 2 results.
|
| 91 |
+
|
| 92 |
+
Args:
|
| 93 |
+
query: The search query."""
|
| 94 |
+
search_docs = WikipediaLoader(query=query, load_max_docs=2).load()
|
| 95 |
+
formatted_search_docs = "\n\n---\n\n".join(
|
| 96 |
+
[
|
| 97 |
+
f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
|
| 98 |
+
for doc in search_docs
|
| 99 |
])
|
| 100 |
+
return {"wiki_results": formatted_search_docs}
|
|
|
|
| 101 |
|
| 102 |
@tool
|
| 103 |
+
def web_search(query: str) -> str:
|
| 104 |
+
"""Search Tavily for a query and return maximum 3 results.
|
| 105 |
+
|
| 106 |
+
Args:
|
| 107 |
+
query: The search query."""
|
| 108 |
+
search_docs = TavilySearchResults(max_results=3).invoke(query=query)
|
| 109 |
+
formatted_search_docs = "\n\n---\n\n".join(
|
| 110 |
+
[
|
| 111 |
+
f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
|
| 112 |
+
for doc in search_docs
|
| 113 |
])
|
| 114 |
+
return {"web_results": formatted_search_docs}
|
|
|
|
| 115 |
|
| 116 |
@tool
|
| 117 |
+
def arvix_search(query: str) -> str:
|
| 118 |
+
"""Search Arxiv for a query and return maximum 3 result.
|
| 119 |
+
|
| 120 |
+
Args:
|
| 121 |
+
query: The search query."""
|
| 122 |
+
search_docs = ArxivLoader(query=query, load_max_docs=3).load()
|
| 123 |
+
formatted_search_docs = "\n\n---\n\n".join(
|
| 124 |
+
[
|
| 125 |
+
f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
|
| 126 |
+
for doc in search_docs
|
| 127 |
+
])
|
| 128 |
+
return {"arvix_results": formatted_search_docs}
|
| 129 |
|
| 130 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 131 |
|
| 132 |
+
# load the system prompt from the file
|
| 133 |
+
with open("system_prompt.txt", "r", encoding="utf-8") as f:
|
| 134 |
+
system_prompt = f.read()
|
|
|
|
|
|
|
|
|
|
| 135 |
|
| 136 |
+
# System message
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
sys_msg = SystemMessage(content=system_prompt)
|
| 138 |
|
| 139 |
+
# build a retriever
|
| 140 |
+
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2") # dim=768
|
| 141 |
+
supabase: Client = create_client(
|
| 142 |
+
os.environ.get("SUPABASE_URL"),
|
| 143 |
+
os.environ.get("SUPABASE_SERVICE_KEY"))
|
| 144 |
+
vector_store = SupabaseVectorStore(
|
| 145 |
+
client=supabase,
|
| 146 |
+
embedding= embeddings,
|
| 147 |
+
table_name="documents",
|
| 148 |
+
query_name="match_documents_langchain",
|
| 149 |
+
)
|
| 150 |
+
create_retriever_tool = create_retriever_tool(
|
| 151 |
+
retriever=vector_store.as_retriever(),
|
| 152 |
+
name="Question Search",
|
| 153 |
+
description="A tool to retrieve similar questions from a vector store.",
|
| 154 |
+
)
|
| 155 |
|
|
|
|
| 156 |
tools = [
|
| 157 |
+
multiply,
|
| 158 |
+
add,
|
| 159 |
+
subtract,
|
| 160 |
+
divide,
|
| 161 |
+
modulus,
|
| 162 |
+
wiki_search,
|
| 163 |
+
web_search,
|
| 164 |
+
arvix_search,
|
| 165 |
]
|
| 166 |
|
| 167 |
+
# Build graph function
|
| 168 |
+
def build_graph(provider: str = "google"):
|
| 169 |
+
"""Build the graph"""
|
| 170 |
+
# Load environment variables from .env file
|
| 171 |
+
if provider == "google":
|
| 172 |
+
# Google Gemini
|
| 173 |
+
llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)
|
| 174 |
+
elif provider == "huggingface":
|
| 175 |
+
# TODO: Add huggingface endpoint
|
| 176 |
+
llm = ChatHuggingFace(
|
| 177 |
+
llm=HuggingFaceEndpoint(
|
| 178 |
+
url="https://api-inference.huggingface.co/models/Meta-DeepLearning/llama-2-7b-chat-hf",
|
| 179 |
+
temperature=0,
|
| 180 |
+
),
|
| 181 |
)
|
| 182 |
+
else:
|
| 183 |
+
raise ValueError("Invalid provider. Choose 'google', or 'huggingface'.")
|
| 184 |
+
# Bind tools to LLM
|
| 185 |
+
llm_with_tools = llm.bind_tools(tools)
|
| 186 |
+
|
| 187 |
+
# Node
|
| 188 |
+
def assistant(state: MessagesState):
|
| 189 |
+
"""Assistant node"""
|
| 190 |
+
return {"messages": [llm_with_tools.invoke(state["messages"])]}
|
| 191 |
+
|
| 192 |
+
def retriever(state: MessagesState):
|
| 193 |
+
"""Retriever node"""
|
| 194 |
+
similar_question = vector_store.similarity_search(state["messages"][0].content)
|
| 195 |
+
example_msg = HumanMessage(
|
| 196 |
+
content=f"Here I provide a similar question and answer for reference: \n\n{similar_question[0].page_content}",
|
| 197 |
+
)
|
| 198 |
+
return {"messages": [sys_msg] + state["messages"] + [example_msg]}
|
| 199 |
+
|
| 200 |
+
builder = StateGraph(MessagesState)
|
| 201 |
+
builder.add_node("retriever", retriever)
|
| 202 |
+
builder.add_node("assistant", assistant)
|
| 203 |
+
builder.add_node("tools", ToolNode(tools))
|
| 204 |
+
builder.add_edge(START, "retriever")
|
| 205 |
+
builder.add_edge("retriever", "assistant")
|
| 206 |
+
builder.add_conditional_edges(
|
| 207 |
+
"assistant",
|
| 208 |
+
tools_condition,
|
| 209 |
)
|
| 210 |
+
builder.add_edge("tools", "assistant")
|
| 211 |
|
| 212 |
+
# Compile graph
|
| 213 |
+
return builder.compile()
|
| 214 |
+
|
| 215 |
+
|
| 216 |
+
|
| 217 |
+
# Tool Definitions
|
| 218 |
+
# @tool
|
| 219 |
+
# def multiply(a: int, b: int) -> int:
|
| 220 |
+
# """Multiply two numbers."""
|
| 221 |
+
# return a * b
|
| 222 |
+
|
| 223 |
+
# @tool
|
| 224 |
+
# def add(a: int, b: int) -> int:
|
| 225 |
+
# """Add two numbers."""
|
| 226 |
+
# return a + b
|
| 227 |
+
|
| 228 |
+
# @tool
|
| 229 |
+
# def modulus(a: int, b: int) -> int:
|
| 230 |
+
# """Get the modulus of two numbers."""
|
| 231 |
+
# return a % b
|
| 232 |
+
|
| 233 |
+
# @tool
|
| 234 |
+
# def subtract(a: int, b: int) -> int:
|
| 235 |
+
# """Subtract two numbers."""
|
| 236 |
+
# return a - b
|
| 237 |
+
|
| 238 |
+
# @tool
|
| 239 |
+
# def divide(a: int, b: int) -> int:
|
| 240 |
+
# """Divide two numbers."""
|
| 241 |
+
# if b == 0:
|
| 242 |
+
# raise ValueError("Cannot divide by zero.")
|
| 243 |
+
# return a / b
|
| 244 |
+
|
| 245 |
+
# @tool
|
| 246 |
+
# def arvix_search(query: str) -> str:
|
| 247 |
+
# """Search Arxiv for a query and return maximum 3 results."""
|
| 248 |
+
# try:
|
| 249 |
+
# search_docs = ArxivLoader(query=query, load_max_docs=3).load()
|
| 250 |
+
# formatted_search_docs = "\n\n---\n\n".join(
|
| 251 |
+
# [f'<Document source="{doc.metadata["source"]}"/>\n{doc.page_content[:1000]}\n</Document>'
|
| 252 |
+
# for doc in search_docs])
|
| 253 |
+
# return {"arvix_results": formatted_search_docs}
|
| 254 |
+
# except Exception as e:
|
| 255 |
+
# return {"arvix_results": f"Error: {str(e)}"}
|
| 256 |
+
|
| 257 |
+
# @tool
|
| 258 |
+
# def execute_python(code: str) -> str:
|
| 259 |
+
# """Execute Python code securely and return results. Handles calculations and data analysis."""
|
| 260 |
+
# try:
|
| 261 |
+
# parsed = ast.parse(code)
|
| 262 |
+
# if any(isinstance(node, (ast.Import, ast.ImportFrom)) for node in parsed.body):
|
| 263 |
+
# return "Cannot import modules for security reasons"
|
| 264 |
+
|
| 265 |
+
# restricted_builtins = {'open', 'eval', 'exec', '__import__'}
|
| 266 |
+
# for node in ast.walk(parsed):
|
| 267 |
+
# if isinstance(node, ast.Name) and node.id in restricted_builtins:
|
| 268 |
+
# return "Restricted function used"
|
| 269 |
+
|
| 270 |
+
# return str(eval(code, {"__builtins__": {}}, {}))
|
| 271 |
+
# except Exception as e:
|
| 272 |
+
# return f"Execution error: {str(e)}"
|
| 273 |
+
|
| 274 |
+
# @tool
|
| 275 |
+
# def process_file(file_path: str) -> Dict[str, Any]:
|
| 276 |
+
# """Process uploaded files (Excel, CSV, TXT) and extract structured data."""
|
| 277 |
+
# try:
|
| 278 |
+
# if file_path.endswith(('.xlsx', '.xls')):
|
| 279 |
+
# df = pd.read_excel(file_path)
|
| 280 |
+
# return {"data": df.head(10).to_dict(), "summary": df.describe().to_dict()}
|
| 281 |
+
# elif file_path.endswith('.csv'):
|
| 282 |
+
# df = pd.read_csv(file_path)
|
| 283 |
+
# return {"data": df.head(10).to_dict(), "summary": df.describe().to_dict()}
|
| 284 |
+
# elif file_path.endswith('.txt'):
|
| 285 |
+
# with open(file_path, 'r') as f:
|
| 286 |
+
# content = f.read(2000)
|
| 287 |
+
# return {"content": content}
|
| 288 |
+
# return {"error": "Unsupported file format"}
|
| 289 |
+
# except Exception as e:
|
| 290 |
+
# return {"error": str(e)}
|
| 291 |
+
|
| 292 |
+
# @tool
|
| 293 |
+
# def enhanced_wiki_search(query: str) -> str:
|
| 294 |
+
# """Access full Wikipedia articles with history and specific versions."""
|
| 295 |
+
# try:
|
| 296 |
+
# loader = WikipediaLoader(query=query, load_max_docs=2, doc_content_chars_max=4000)
|
| 297 |
+
# docs = loader.load()
|
| 298 |
+
# return "\n\n".join([
|
| 299 |
+
# f"Title: {doc.metadata['title']}\n"
|
| 300 |
+
# f"URL: {doc.metadata['source']}\n"
|
| 301 |
+
# f"Content: {doc.page_content[:3000]}..."
|
| 302 |
+
# for doc in docs
|
| 303 |
+
# ])
|
| 304 |
+
# except Exception as e:
|
| 305 |
+
# return f"Wikipedia Error: {str(e)}"
|
| 306 |
+
|
| 307 |
+
# @tool
|
| 308 |
+
# def media_search(query: str) -> str:
|
| 309 |
+
# """Specialized search for media content (videos, images, celebrities)."""
|
| 310 |
+
# try:
|
| 311 |
+
# search = DuckDuckGoSearchResults(max_results=5)
|
| 312 |
+
# results = search.invoke(f"site:imdb.com OR site:youtube.com {query}")
|
| 313 |
+
# return "\n".join([
|
| 314 |
+
# f"Source: {res['link']}\nSnippet: {res['snippet']}"
|
| 315 |
+
# for res in results[:3]
|
| 316 |
+
# ])
|
| 317 |
+
# except Exception as e:
|
| 318 |
+
# return f"Media Search Error: {str(e)}"
|
| 319 |
+
|
| 320 |
+
# @tool
|
| 321 |
+
# def academic_search(query: str) -> str:
|
| 322 |
+
# """Search academic databases and educational resources."""
|
| 323 |
+
# try:
|
| 324 |
+
# arxiv_docs = ArxivLoader(query=query, load_max_docs=2).load()
|
| 325 |
+
# web_docs = DuckDuckGoSearchResults(max_results=3).invoke(f"filetype:pdf {query}")
|
| 326 |
+
# return f"Arxiv Results:\n{arxiv_docs[0].page_content[:1000]}\n\nWeb Results:\n{web_docs[0]['snippet']}"
|
| 327 |
+
# except Exception as e:
|
| 328 |
+
# return f"Academic Search Error: {str(e)}"
|
| 329 |
+
|
| 330 |
+
|
| 331 |
+
# @tool
|
| 332 |
+
# def web_search(query: str) -> str:
|
| 333 |
+
# """Search DuckDuckGo for a query and return maximum 3 results."""
|
| 334 |
+
# try:
|
| 335 |
+
# search = DuckDuckGoSearchResults(max_results=3)
|
| 336 |
+
# search_docs = search.invoke(query)
|
| 337 |
+
# formatted_search_docs = "\n\n---\n\n".join(
|
| 338 |
+
# [f'<Document source="{doc["link"]}"/>\n{doc["snippet"]}\n</Document>'
|
| 339 |
+
# for doc in search_docs])
|
| 340 |
+
# return {"web_results": formatted_search_docs}
|
| 341 |
+
# except Exception as e:
|
| 342 |
+
# return {"web_results": f"Error: {str(e)}"}
|
| 343 |
+
|
| 344 |
+
# @tool
|
| 345 |
+
# def summarize_text(text: str) -> str:
|
| 346 |
+
# """Summarize a long text into key points."""
|
| 347 |
+
# if not text or len(text) < 100:
|
| 348 |
+
# return "Text too short to summarize."
|
| 349 |
+
# return f"Summary of the provided text with key points."
|
| 350 |
+
|
| 351 |
+
# @tool
|
| 352 |
+
# def parse_query(query: str) -> dict:
|
| 353 |
+
# """Parse a complex query into its key components for better search."""
|
| 354 |
+
# parts = query.split()
|
| 355 |
+
# return {
|
| 356 |
+
# "main_topic": parts[0] if parts else "",
|
| 357 |
+
# "subtopics": parts[1:3] if len(parts) > 1 else [],
|
| 358 |
+
# "context": " ".join(parts[3:]) if len(parts) > 3 else ""
|
| 359 |
+
# }
|
| 360 |
+
|
| 361 |
+
# # System Prompt Setup
|
| 362 |
+
# system_prompt = """You are a POWERFUL assistant REQUIRED to answer ALL questions using available tools.
|
| 363 |
+
# STRICT RULES:
|
| 364 |
+
# 1. NEVER say you can't answer - ALWAYS use tools
|
| 365 |
+
# 2. Combine information from multiple tools when needed
|
| 366 |
+
# 3. For calculations, use execute_python
|
| 367 |
+
# 4. For files, use process_file
|
| 368 |
+
# 5. For media/celebrities, use media_search
|
| 369 |
+
# 6. For academic content, use academic_search
|
| 370 |
+
# 7. ALWAYS format final answer as: FINAL ANSWER: [your answer]
|
| 371 |
+
|
| 372 |
+
# AVAILABLE TOOLS:
|
| 373 |
+
# - execute_python: Math/code execution
|
| 374 |
+
# - process_file: Analyze uploaded files
|
| 375 |
+
# - enhanced_wiki_search: Full Wikipedia access
|
| 376 |
+
# - media_search: Videos/images/celebrities
|
| 377 |
+
# - academic_search: Textbooks/papers
|
| 378 |
+
# - web_search: General web search
|
| 379 |
+
# - vector_store: Previous knowledge
|
| 380 |
+
|
| 381 |
+
# YOU MUST USE THESE TOOLS TO ANSWER ALL QUESTIONS!"""
|
| 382 |
+
# sys_msg = SystemMessage(content=system_prompt)
|
| 383 |
+
|
| 384 |
+
# # Vector Store Setup
|
| 385 |
+
# try:
|
| 386 |
+
# embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
|
| 387 |
+
# vector_store = Chroma(
|
| 388 |
+
# collection_name="documents",
|
| 389 |
+
# embedding_function=embeddings,
|
| 390 |
+
# persist_directory="./chroma_db"
|
| 391 |
+
# )
|
| 392 |
+
# except Exception as e:
|
| 393 |
+
# print(f"Error initializing vector store: {e}")
|
| 394 |
+
# vector_store = None
|
| 395 |
+
|
| 396 |
+
# # Tool Configuration
|
| 397 |
+
# tools = [
|
| 398 |
+
# multiply, add, subtract, divide, modulus,
|
| 399 |
+
# enhanced_wiki_search, media_search, web_search, arvix_search,
|
| 400 |
+
# academic_search, summarize_text, parse_query, DuckDuckGoSearchResults(max_results=5)
|
| 401 |
+
# ]
|
| 402 |
+
|
| 403 |
+
# if vector_store:
|
| 404 |
+
# tools.append(
|
| 405 |
+
# create_retriever_tool(
|
| 406 |
+
# vector_store.as_retriever(),
|
| 407 |
+
# name="Question Search",
|
| 408 |
+
# description="Retrieves similar questions from vector store"
|
| 409 |
+
# )
|
| 410 |
+
# )
|
| 411 |
+
|
| 412 |
+
# # Model Configuration
|
| 413 |
+
# MODEL_REGISTRY = {
|
| 414 |
+
# "gemini-2.0-flash": {
|
| 415 |
+
# "provider": "google",
|
| 416 |
+
# "model": "gemini-2.0-flash",
|
| 417 |
+
# "temperature": 0.2,
|
| 418 |
+
# "max_tokens": 2048
|
| 419 |
+
# },
|
| 420 |
+
# "gemini-1.5-flash": {
|
| 421 |
+
# "provider": "google",
|
| 422 |
+
# "model": "gemini-1.5-flash",
|
| 423 |
+
# "temperature": 0.2,
|
| 424 |
+
# "max_tokens": 2048
|
| 425 |
+
# },
|
| 426 |
+
# "kimi-vl-a3b-thinking": {
|
| 427 |
+
# "provider": "openrouter",
|
| 428 |
+
# "model": "moonshotai/kimi-vl-a3b-thinking:free",
|
| 429 |
+
# "temperature": 0.2,
|
| 430 |
+
# "max_tokens": 2048
|
| 431 |
+
# }
|
| 432 |
+
# }
|
| 433 |
+
|
| 434 |
+
# def get_llm(model_name: str = "gemini-2.0-flash"):
|
| 435 |
+
# """Initialize LLM with error handling"""
|
| 436 |
+
# config = MODEL_REGISTRY.get(model_name, MODEL_REGISTRY["gemini-2.0-flash"])
|
| 437 |
+
# provider = config.get("provider", "google")
|
| 438 |
|
| 439 |
+
# try:
|
| 440 |
+
# if provider == "google":
|
| 441 |
+
# if not GOOGLE_API_KEY:
|
| 442 |
+
# print(f"Error initializing {model_name}: GOOGLE_API_KEY not found")
|
| 443 |
+
# return None
|
| 444 |
+
# return ChatGoogleGenerativeAI(
|
| 445 |
+
# model=config["model"],
|
| 446 |
+
# temperature=config["temperature"],
|
| 447 |
+
# max_output_tokens=config["max_tokens"],
|
| 448 |
+
# convert_system_message_to_human=True
|
| 449 |
+
# )
|
| 450 |
+
# elif provider == "openrouter":
|
| 451 |
+
# if not OPENROUTER_API_KEY:
|
| 452 |
+
# print(f"Error initializing {model_name}: OPENROUTER_API_KEY not found")
|
| 453 |
+
# return None
|
| 454 |
+
# return ChatOpenAI(
|
| 455 |
+
# model=config["model"],
|
| 456 |
+
# temperature=config["temperature"],
|
| 457 |
+
# max_tokens=config["max_tokens"],
|
| 458 |
+
# openai_api_key=OPENROUTER_API_KEY,
|
| 459 |
+
# openai_api_base="https://openrouter.ai/api/v1",
|
| 460 |
+
# model_kwargs={
|
| 461 |
+
# "headers": {
|
| 462 |
+
# "HTTP-Referer": "https://your-site.com",
|
| 463 |
+
# "X-Title": "Agent Evaluation"
|
| 464 |
+
# }
|
| 465 |
+
# }
|
| 466 |
+
# )
|
| 467 |
+
# else:
|
| 468 |
+
# print(f"Unknown provider {provider} for model {model_name}")
|
| 469 |
+
# return None
|
| 470 |
+
# except Exception as e:
|
| 471 |
+
# print(f"Error initializing {model_name}: {e}")
|
| 472 |
+
# return None
|
| 473 |
+
|
| 474 |
+
# # Graph Builder
|
| 475 |
+
# def build_graph():
|
| 476 |
+
# """Build LangGraph agent workflow"""
|
| 477 |
+
# primary_llm = get_llm("gemini-2.0-flash")
|
| 478 |
+
# fallback_llm = get_llm("gemini-1.5-flash")
|
| 479 |
+
# kimi_llm = get_llm("kimi-vl-a3b-thinking")
|
| 480 |
|
| 481 |
+
# llms = [llm for llm in [primary_llm, fallback_llm, kimi_llm] if llm is not None]
|
| 482 |
|
| 483 |
+
# if not llms:
|
| 484 |
+
# raise RuntimeError("Failed to initialize any LLM")
|
| 485 |
|
| 486 |
+
# current_llm_index = 0
|
| 487 |
|
| 488 |
+
# def assistant(state: MessagesState):
|
| 489 |
+
# nonlocal current_llm_index
|
| 490 |
+
# for attempt in range(len(llms)):
|
| 491 |
+
# try:
|
| 492 |
+
# llm = llms[current_llm_index]
|
| 493 |
+
# llm_with_tools = llm.bind_tools(tools)
|
| 494 |
|
| 495 |
+
# messages = state["messages"].copy()
|
| 496 |
+
# if len(messages) > 0 and isinstance(messages[0], HumanMessage):
|
| 497 |
+
# tool_instruction = HumanMessage(content="Use available tools to answer.")
|
| 498 |
+
# messages.append(tool_instruction)
|
| 499 |
|
| 500 |
+
# response = llm_with_tools.invoke(messages)
|
| 501 |
+
# current_llm_index = (current_llm_index + 1) % len(llms)
|
| 502 |
+
# return {"messages": [response]}
|
| 503 |
+
# except Exception as e:
|
| 504 |
+
# print(f"Model {llms[current_llm_index]} failed: {e}")
|
| 505 |
+
# current_llm_index = (current_llm_index + 1) % len(llms)
|
| 506 |
+
# if attempt == len(llms) - 1:
|
| 507 |
+
# error_msg = HumanMessage(content=f"All models failed: {str(e)}")
|
| 508 |
+
# return {"messages": [error_msg]}
|
| 509 |
+
|
| 510 |
+
# def retriever(state: MessagesState):
|
| 511 |
+
# try:
|
| 512 |
+
# if vector_store:
|
| 513 |
+
# similar_questions = vector_store.similarity_search(
|
| 514 |
+
# state["messages"][0].content,
|
| 515 |
+
# k=1
|
| 516 |
+
# )
|
| 517 |
+
# example_content = "Similar question reference: \n\n" + \
|
| 518 |
+
# (similar_questions[0].page_content if similar_questions
|
| 519 |
+
# else "No similar questions found")
|
| 520 |
+
# else:
|
| 521 |
+
# example_content = "Vector store not available"
|
| 522 |
|
| 523 |
+
# return {"messages": [sys_msg] + state["messages"] + [HumanMessage(content=example_content)]}
|
| 524 |
+
# except Exception as e:
|
| 525 |
+
# error_msg = HumanMessage(content=f"Retrieval error: {str(e)}")
|
| 526 |
+
# return {"messages": [error_msg]}
|
| 527 |
|
| 528 |
+
# builder = StateGraph(MessagesState)
|
| 529 |
+
# builder.add_node("retriever", retriever)
|
| 530 |
+
# builder.add_node("assistant", assistant)
|
| 531 |
+
# builder.add_node("tools", ToolNode(tools))
|
| 532 |
|
| 533 |
+
# builder.add_edge(START, "retriever")
|
| 534 |
+
# builder.add_edge("retriever", "assistant")
|
| 535 |
+
# builder.add_conditional_edges("assistant", tools_condition)
|
| 536 |
+
# builder.add_edge("tools", "assistant")
|
| 537 |
|
| 538 |
+
# return builder.compile()
|
| 539 |
|
| 540 |
# Agent Class
|
| 541 |
class BasicAgent:
|