Update app.py
Browse files
app.py
CHANGED
|
@@ -1,25 +1,26 @@
|
|
| 1 |
-
from smolagents import CodeAgent,
|
| 2 |
import datetime
|
| 3 |
-
import
|
| 4 |
-
import pytz # Used in the commented-out example tool
|
| 5 |
import yaml
|
| 6 |
import os
|
| 7 |
import tempfile
|
| 8 |
-
|
| 9 |
-
from tools.visit_webpage import VisitWebpageTool # Assuming this exists in ./tools/visit_webpage.py
|
| 10 |
-
from tools.web_search import DuckDuckGoSearchTool
|
| 11 |
-
|
| 12 |
-
import gradio as gr
|
| 13 |
-
import json # Not used directly in the provided snippet, but can be common
|
| 14 |
-
|
| 15 |
from typing import Dict, List, Optional, Union, Any
|
| 16 |
-
import re # Added for regex in Gradio UI if not already globally available
|
| 17 |
|
| 18 |
-
|
| 19 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
-
|
|
|
|
| 22 |
|
|
|
|
| 23 |
|
| 24 |
@tool
|
| 25 |
def create_document(text: str, format: str = "docx") -> str:
|
|
@@ -30,7 +31,7 @@ def create_document(text: str, format: str = "docx") -> str:
|
|
| 30 |
"""
|
| 31 |
try:
|
| 32 |
temp_dir = tempfile.mkdtemp()
|
| 33 |
-
file_name = "generated_document"
|
| 34 |
|
| 35 |
if format.lower() == "txt":
|
| 36 |
path = os.path.join(temp_dir, f"{file_name}.txt")
|
|
@@ -75,17 +76,14 @@ def create_document(text: str, format: str = "docx") -> str:
|
|
| 75 |
f"Please install it (e.g., 'pip install docx2pdf'). "
|
| 76 |
f"Document saved as DOCX instead at: {docx_path}")
|
| 77 |
print(err_msg)
|
| 78 |
-
return err_msg
|
| 79 |
except Exception as e_pdf:
|
| 80 |
err_msg = f"Error converting DOCX to PDF: {str(e_pdf)}. Document saved as DOCX at: {docx_path}"
|
| 81 |
print(err_msg)
|
| 82 |
-
return err_msg
|
| 83 |
-
|
| 84 |
return docx_path
|
| 85 |
-
|
| 86 |
else:
|
| 87 |
return f"Error: Unsupported format '{format}'. Supported formats are 'docx', 'pdf', 'txt'."
|
| 88 |
-
|
| 89 |
except Exception as e:
|
| 90 |
print(f"General error in create_document: {str(e)}")
|
| 91 |
return f"Error creating document: {str(e)}"
|
|
@@ -110,40 +108,120 @@ def get_file_download_link(file_path: str) -> str:
|
|
| 110 |
msg = f"File '{os.path.basename(file_path)}' is ready for download (type: {mime_type})."
|
| 111 |
print(f"get_file_download_link: {msg}")
|
| 112 |
return msg
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
final_answer = FinalAnswerTool()
|
| 115 |
-
web_search=DuckDuckGoSearchTool()
|
| 116 |
-
visit_webpage=VisitWebpageTool()
|
| 117 |
|
| 118 |
-
# Load image generation tool from Hub
|
| 119 |
try:
|
| 120 |
-
image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
|
| 121 |
-
# Common alternative: "agents-course/text-to-image" (uses a smaller SD model)
|
| 122 |
-
# image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
|
| 123 |
print(f"Successfully loaded image generation tool. Name: {getattr(image_generation_tool, 'name', 'N/A')}")
|
| 124 |
except Exception as e:
|
| 125 |
-
print(f"Error loading image generation tool: {e}")
|
| 126 |
-
|
| 127 |
-
image_generation_tool = None # Set to None if loading fails
|
| 128 |
|
| 129 |
# --- Configure LLM Model ---
|
| 130 |
-
# Ensure GEMINI_KEY is set in your environment variables
|
| 131 |
if not os.getenv("GEMINI_KEY"):
|
| 132 |
-
print("CRITICAL: GEMINI_KEY environment variable not set.
|
| 133 |
# exit(1) # Or handle gracefully
|
| 134 |
|
| 135 |
model = LiteLLMModel(
|
| 136 |
-
model_id="gemini/gemini-1.5-flash-latest",
|
| 137 |
api_key=os.getenv("GEMINI_KEY"),
|
| 138 |
-
max_tokens=4096,
|
| 139 |
-
temperature=0.6
|
| 140 |
)
|
| 141 |
|
| 142 |
# --- Load Prompts ---
|
| 143 |
prompts_file = "prompts.yaml"
|
| 144 |
if not os.path.exists(prompts_file):
|
| 145 |
print(f"Warning: '{prompts_file}' not found. Using default agent prompts.")
|
| 146 |
-
prompt_templates = None
|
| 147 |
else:
|
| 148 |
with open(prompts_file, 'r') as stream:
|
| 149 |
prompt_templates = yaml.safe_load(stream)
|
|
@@ -153,8 +231,9 @@ agent_tools = [
|
|
| 153 |
final_answer,
|
| 154 |
web_search,
|
| 155 |
visit_webpage,
|
| 156 |
-
create_document,
|
| 157 |
-
get_file_download_link,
|
|
|
|
| 158 |
]
|
| 159 |
|
| 160 |
if image_generation_tool:
|
|
@@ -165,16 +244,18 @@ else:
|
|
| 165 |
agent = CodeAgent(
|
| 166 |
model=model,
|
| 167 |
tools=agent_tools,
|
| 168 |
-
max_steps=10,
|
| 169 |
-
verbosity_level=1,
|
| 170 |
prompt_templates=prompt_templates
|
| 171 |
)
|
| 172 |
|
| 173 |
# --- Main Application Entry Point ---
|
| 174 |
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
|
|
| 175 |
print("Starting Gradio UI...")
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
ui = GradioUI(agent, file_upload_folder="uploaded_files") # Enable file uploads
|
| 180 |
-
ui.launch() # share=False is default and safer for Spaces
|
|
|
|
| 1 |
+
from smolagents import CodeAgent, load_tool, tool, LiteLLMModel
|
| 2 |
import datetime
|
| 3 |
+
import pytz
|
|
|
|
| 4 |
import yaml
|
| 5 |
import os
|
| 6 |
import tempfile
|
| 7 |
+
import re
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
from typing import Dict, List, Optional, Union, Any
|
|
|
|
| 9 |
|
| 10 |
+
# --- Tool Specific Imports ---
|
| 11 |
+
from newsapi import NewsApiClient # For News Tool
|
| 12 |
+
|
| 13 |
+
# --- Assuming your tools are structured like this ---
|
| 14 |
+
# If these are classes, they should be instantiated.
|
| 15 |
+
# If they are @tool decorated functions, they are used directly.
|
| 16 |
+
from tools.final_answer import FinalAnswerTool
|
| 17 |
+
from tools.visit_webpage import VisitWebpageTool
|
| 18 |
+
from tools.web_search import DuckDuckGoSearchTool
|
| 19 |
|
| 20 |
+
# --- Gradio UI ---
|
| 21 |
+
# from Gradio_UI import GradioUI # Assuming Gradio_UI.py is in the same directory
|
| 22 |
|
| 23 |
+
# --- @tool Decorated Functions (Custom Tools) ---
|
| 24 |
|
| 25 |
@tool
|
| 26 |
def create_document(text: str, format: str = "docx") -> str:
|
|
|
|
| 31 |
"""
|
| 32 |
try:
|
| 33 |
temp_dir = tempfile.mkdtemp()
|
| 34 |
+
file_name = "generated_document"
|
| 35 |
|
| 36 |
if format.lower() == "txt":
|
| 37 |
path = os.path.join(temp_dir, f"{file_name}.txt")
|
|
|
|
| 76 |
f"Please install it (e.g., 'pip install docx2pdf'). "
|
| 77 |
f"Document saved as DOCX instead at: {docx_path}")
|
| 78 |
print(err_msg)
|
| 79 |
+
return err_msg
|
| 80 |
except Exception as e_pdf:
|
| 81 |
err_msg = f"Error converting DOCX to PDF: {str(e_pdf)}. Document saved as DOCX at: {docx_path}"
|
| 82 |
print(err_msg)
|
| 83 |
+
return err_msg
|
|
|
|
| 84 |
return docx_path
|
|
|
|
| 85 |
else:
|
| 86 |
return f"Error: Unsupported format '{format}'. Supported formats are 'docx', 'pdf', 'txt'."
|
|
|
|
| 87 |
except Exception as e:
|
| 88 |
print(f"General error in create_document: {str(e)}")
|
| 89 |
return f"Error creating document: {str(e)}"
|
|
|
|
| 108 |
msg = f"File '{os.path.basename(file_path)}' is ready for download (type: {mime_type})."
|
| 109 |
print(f"get_file_download_link: {msg}")
|
| 110 |
return msg
|
| 111 |
+
|
| 112 |
+
@tool
|
| 113 |
+
def get_latest_news(query: Optional[str] = None,
|
| 114 |
+
category: Optional[str] = None,
|
| 115 |
+
country: Optional[str] = "us",
|
| 116 |
+
language: Optional[str] = "en",
|
| 117 |
+
page_size: int = 5) -> str:
|
| 118 |
+
"""
|
| 119 |
+
Fetches the latest news headlines.
|
| 120 |
+
You can specify a query, category, country, and language.
|
| 121 |
+
|
| 122 |
+
Args:
|
| 123 |
+
query: Keywords or a phrase to search for in the news articles. (e.g., "Tesla stock")
|
| 124 |
+
category: The category: business, entertainment, general, health, science, sports, technology.
|
| 125 |
+
country: The 2-letter ISO 3166-1 code of the country (e.g., 'us', 'gb'). Default 'us'.
|
| 126 |
+
language: The 2-letter ISO 639-1 code of the language (e.g., 'en', 'es'). Default 'en'.
|
| 127 |
+
page_size: Number of results (max 100, default 5).
|
| 128 |
+
"""
|
| 129 |
+
api_key = os.getenv("NEWS_API_KEY")
|
| 130 |
+
if not api_key:
|
| 131 |
+
return "ERROR: News API key (NEWS_API_KEY) is not set in environment variables."
|
| 132 |
+
|
| 133 |
+
newsapi = NewsApiClient(api_key=api_key)
|
| 134 |
+
print(f"DEBUG NewsTool: query='{query}', category='{category}', country='{country}', lang='{language}', size={page_size}")
|
| 135 |
+
|
| 136 |
+
try:
|
| 137 |
+
if query:
|
| 138 |
+
print(f"Fetching news with query: '{query}'")
|
| 139 |
+
top_headlines = newsapi.get_everything(q=query,
|
| 140 |
+
language=language,
|
| 141 |
+
sort_by='publishedAt',
|
| 142 |
+
page_size=page_size)
|
| 143 |
+
elif category or country != "us": # If category or specific country (other than default for top-headlines)
|
| 144 |
+
print(f"Fetching top headlines for category: '{category}', country: '{country}'")
|
| 145 |
+
top_headlines = newsapi.get_top_headlines(q=None,
|
| 146 |
+
category=category,
|
| 147 |
+
language=language,
|
| 148 |
+
country=country,
|
| 149 |
+
page_size=page_size)
|
| 150 |
+
else: # Default top headlines for US/English if nothing specific is asked
|
| 151 |
+
print(f"Fetching default top headlines for country: '{country}', language: '{language}'")
|
| 152 |
+
top_headlines = newsapi.get_top_headlines(language=language,
|
| 153 |
+
country=country,
|
| 154 |
+
page_size=page_size)
|
| 155 |
|
| 156 |
+
if top_headlines['status'] == 'ok':
|
| 157 |
+
articles = top_headlines['articles']
|
| 158 |
+
if not articles:
|
| 159 |
+
return "No news articles found for the given criteria."
|
| 160 |
+
|
| 161 |
+
formatted_news = "## Latest News:\n\n"
|
| 162 |
+
for i, article in enumerate(articles[:page_size]): # Ensure we respect page_size
|
| 163 |
+
title = article.get('title', 'N/A')
|
| 164 |
+
source_name = article.get('source', {}).get('name', 'N/A')
|
| 165 |
+
description = article.get('description', 'No description available.')
|
| 166 |
+
url = article.get('url', '#')
|
| 167 |
+
published_at_str = article.get('publishedAt', 'N/A')
|
| 168 |
+
|
| 169 |
+
published_at_formatted = published_at_str
|
| 170 |
+
try:
|
| 171 |
+
if published_at_str and published_at_str != 'N/A':
|
| 172 |
+
# Handle 'Z' for UTC timezone correctly
|
| 173 |
+
dt_object = datetime.datetime.fromisoformat(published_at_str.replace('Z', '+00:00'))
|
| 174 |
+
# Format to a more readable string, perhaps without timezone if it's always UTC from API
|
| 175 |
+
published_at_formatted = dt_object.strftime('%Y-%m-%d %H:%M') + " UTC"
|
| 176 |
+
except ValueError:
|
| 177 |
+
pass # Keep original string if parsing fails
|
| 178 |
+
|
| 179 |
+
formatted_news += (
|
| 180 |
+
f"{i+1}. **{title}**\n"
|
| 181 |
+
f" - Source: {source_name}\n"
|
| 182 |
+
f" - Published: {published_at_formatted}\n"
|
| 183 |
+
f" - Description: {description[:200] + '...' if description and len(description) > 200 else (description or '')}\n"
|
| 184 |
+
f" - URL: {url}\n\n"
|
| 185 |
+
)
|
| 186 |
+
return formatted_news.strip()
|
| 187 |
+
else:
|
| 188 |
+
err_msg = f"ERROR: Could not fetch news. API Response: {top_headlines.get('code')} - {top_headlines.get('message')}"
|
| 189 |
+
print(err_msg)
|
| 190 |
+
return err_msg
|
| 191 |
+
|
| 192 |
+
except Exception as e:
|
| 193 |
+
print(f"ERROR: Exception in get_latest_news: {str(e)}")
|
| 194 |
+
return f"ERROR: An exception occurred while fetching news: {str(e)}"
|
| 195 |
+
|
| 196 |
+
# --- Initialize Tools ---
|
| 197 |
final_answer = FinalAnswerTool()
|
| 198 |
+
web_search = DuckDuckGoSearchTool() # Your custom tool from tools/web_search.py
|
| 199 |
+
visit_webpage = VisitWebpageTool()
|
| 200 |
|
|
|
|
| 201 |
try:
|
| 202 |
+
image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
|
|
|
|
|
|
|
| 203 |
print(f"Successfully loaded image generation tool. Name: {getattr(image_generation_tool, 'name', 'N/A')}")
|
| 204 |
except Exception as e:
|
| 205 |
+
print(f"Error loading image generation tool: {e}. Image generation will not be available.")
|
| 206 |
+
image_generation_tool = None
|
|
|
|
| 207 |
|
| 208 |
# --- Configure LLM Model ---
|
|
|
|
| 209 |
if not os.getenv("GEMINI_KEY"):
|
| 210 |
+
print("CRITICAL: GEMINI_KEY environment variable not set. LLM will not work.")
|
| 211 |
# exit(1) # Or handle gracefully
|
| 212 |
|
| 213 |
model = LiteLLMModel(
|
| 214 |
+
model_id="gemini/gemini-1.5-flash-latest",
|
| 215 |
api_key=os.getenv("GEMINI_KEY"),
|
| 216 |
+
max_tokens=4096,
|
| 217 |
+
temperature=0.6
|
| 218 |
)
|
| 219 |
|
| 220 |
# --- Load Prompts ---
|
| 221 |
prompts_file = "prompts.yaml"
|
| 222 |
if not os.path.exists(prompts_file):
|
| 223 |
print(f"Warning: '{prompts_file}' not found. Using default agent prompts.")
|
| 224 |
+
prompt_templates = None
|
| 225 |
else:
|
| 226 |
with open(prompts_file, 'r') as stream:
|
| 227 |
prompt_templates = yaml.safe_load(stream)
|
|
|
|
| 231 |
final_answer,
|
| 232 |
web_search,
|
| 233 |
visit_webpage,
|
| 234 |
+
create_document,
|
| 235 |
+
get_file_download_link,
|
| 236 |
+
get_latest_news, # Added the new news tool
|
| 237 |
]
|
| 238 |
|
| 239 |
if image_generation_tool:
|
|
|
|
| 244 |
agent = CodeAgent(
|
| 245 |
model=model,
|
| 246 |
tools=agent_tools,
|
| 247 |
+
max_steps=10,
|
| 248 |
+
verbosity_level=1,
|
| 249 |
prompt_templates=prompt_templates
|
| 250 |
)
|
| 251 |
|
| 252 |
# --- Main Application Entry Point ---
|
| 253 |
if __name__ == "__main__":
|
| 254 |
+
# Ensure NEWS_API_KEY is set if you want the news tool to work
|
| 255 |
+
if not os.getenv("NEWS_API_KEY"):
|
| 256 |
+
print("Warning: NEWS_API_KEY environment variable not set. News tool will return an error.")
|
| 257 |
+
|
| 258 |
print("Starting Gradio UI...")
|
| 259 |
+
from Gradio_UI import GradioUI # Assuming Gradio_UI.py is in the same directory
|
| 260 |
+
ui = GradioUI(agent) # Removed file_upload_folder argument to disable UI uploads
|
| 261 |
+
ui.launch()
|
|
|
|
|
|