daqc's picture
Update app.py
cd7d11f verified
raw
history blame
47.3 kB
import json
import mimetypes
import os
import re
import shutil
import threading
import uuid
from typing import Optional
from loguru import logger
from datetime import datetime
import gradio as gr
from dotenv import load_dotenv
from huggingface_hub import login, HfApi
from smolagents import (
CodeAgent,
InferenceClientModel,
Tool,
DuckDuckGoSearchTool,
)
from smolagents.agent_types import (
AgentAudio,
AgentImage,
AgentText,
handle_agent_output_types,
)
from smolagents.gradio_ui import stream_to_gradio
from scripts.text_inspector_tool import TextInspectorTool
from scripts.text_web_browser import (
ArchiveSearchTool,
FinderTool,
FindNextTool,
PageDownTool,
PageUpTool,
SimpleTextBrowser,
VisitTool,
)
from scripts.visual_qa import visualizer
from scripts.cvedb_tool import CVEDBTool
from scripts.report_generator import ReportGeneratorTool
from scripts.epss_tool import EpsTool
from scripts.nvd_tool import NvdTool
from scripts.kevin_tool import KevinTool
# web_search = GoogleSearchTool(provider="serper")
web_search = DuckDuckGoSearchTool()
AUTHORIZED_IMPORTS = [
"requests",
"zipfile",
"pandas",
"numpy",
"sympy",
"json",
"bs4",
"pubchempy",
"xml",
"yahoo_finance",
"Bio",
"sklearn",
"scipy",
"pydub",
"PIL",
"chess",
"PyPDF2",
"pptx",
"torch",
"datetime",
"fractions",
"csv",
"plotly",
"plotly.express",
"plotly.graph_objects",
"jinja2",
]
load_dotenv(override=True)
# Only login if HF_TOKEN is available and valid in environment
if os.getenv("HF_TOKEN"):
try:
login(os.getenv("HF_TOKEN"))
logger.info("Successfully logged in with HF_TOKEN from environment")
except Exception as e:
logger.warning(f"Failed to login with HF_TOKEN from environment: {e}")
logger.info("You can still use the application by providing a valid API key in the interface")
# Global session storage for independent user sessions
user_sessions = {}
session_lock = threading.Lock()
append_answer_lock = threading.Lock()
# Initialize browser
browser = SimpleTextBrowser()
# Initialize tools
ti_tool = TextInspectorTool(None, 20000) # Will be updated with session-specific model
# Initialize vulnerability tools
cvedb_tool = CVEDBTool()
epss_tool = EpsTool()
nvd_tool = NvdTool()
kevin_tool = KevinTool()
# Default tools (will be updated with session-specific models)
WEB_TOOLS = [
web_search, # duckduckgo
VisitTool(browser),
PageUpTool(browser),
PageDownTool(browser),
FinderTool(browser),
FindNextTool(browser),
ArchiveSearchTool(browser),
] + ([ti_tool] if ti_tool else []) + [
cvedb_tool, # CVEDB Tool
# report_generator, # Report generation tool - COMMENTED: Only works locally
epss_tool, # EPSS Tool
nvd_tool, # NVD Tool
kevin_tool, # KEVin Tool
]
def validate_hf_api_key(api_key: str) -> tuple[bool, str]:
"""Validate Hugging Face API key by making a test request."""
if not api_key or not api_key.strip():
return False, "❌ API key cannot be empty"
api_key = api_key.strip()
# Basic format validation
if not api_key.startswith("hf_"):
return False, "❌ Invalid API key format. Hugging Face API keys start with 'hf_'"
try:
# Test the API key by making a simple request
api = HfApi(token=api_key)
# Try to get user info to validate the token
user_info = api.whoami()
return True, f"✅ API key validated successfully! Welcome, {user_info.get('name', 'User')}!"
except Exception as e:
return False, f"❌ Invalid API key: {str(e)}"
def create_model_with_api_key(hf_token: str, model_id: str = None) -> InferenceClientModel:
"""Create a model instance with the provided API key."""
if not model_id:
model_id = "Qwen/Qwen2.5-Coder-32B-Instruct"
return InferenceClientModel(
model_id=model_id,
hf_token=hf_token,
temperature=0.1,
max_tokens=4000,
)
def create_tools_with_model(model: InferenceClientModel):
"""Create tools with the provided model."""
# Update text inspector tool with the model
ti_tool = TextInspectorTool(model, 20000)
return [
web_search, # duckduckgo
VisitTool(browser),
PageUpTool(browser),
PageDownTool(browser),
FinderTool(browser),
FindNextTool(browser),
ArchiveSearchTool(browser),
] + ([ti_tool] if ti_tool else []) + [
cvedb_tool, # CVEDB Tool
# report_generator, # Report generation tool - COMMENTED: Only works locally
epss_tool, # EPSS Tool
nvd_tool, # NVD Tool
kevin_tool, # KEVin Tool
]
# Agent creation in a factory function
def create_agent(hf_token: str = None, model_id: str = None, max_steps: int = 10):
"""Creates a fresh agent instance for each session"""
if not hf_token:
raise ValueError("A valid Hugging Face API key is required to create an agent.")
logger.info(f"Creating agent with token: {hf_token[:10]}...")
# Use session-specific model with HF_TOKEN
model = create_model_with_api_key(hf_token, model_id)
tools = create_tools_with_model(model)
agent = CodeAgent(
model=model,
tools=[visualizer] + tools,
max_steps=max_steps,
verbosity_level=1,
additional_authorized_imports=AUTHORIZED_IMPORTS,
planning_interval=4,
)
logger.info("Agent created successfully")
return agent
# Document inspection tool will be created per session
document_inspection_tool = None
def get_user_session(request: gr.Request) -> str:
"""Get or create a unique session ID for the user."""
if not request:
return str(uuid.uuid4())
# Try to get session from headers or create new one
session_id = request.headers.get("x-session-id")
if not session_id:
session_id = str(uuid.uuid4())
return session_id
def get_session_data(session_id: str) -> dict:
"""Get session data for a specific user."""
with session_lock:
if session_id not in user_sessions:
user_sessions[session_id] = {
"hf_token": None,
"agent": None,
"max_steps": 10,
"created_at": datetime.now()
}
return user_sessions[session_id]
def clear_session_data(session_id: str):
"""Clear session data for a specific user."""
with session_lock:
if session_id in user_sessions:
# Clear sensitive data
user_sessions[session_id]["hf_token"] = None
user_sessions[session_id]["agent"] = None
logger.info(f"Session {session_id[:8]}... cleared")
def cleanup_old_sessions():
"""Clean up sessions older than 1 hour."""
with session_lock:
current_time = datetime.now()
sessions_to_remove = []
for session_id, session_data in user_sessions.items():
if session_data.get("created_at"):
time_diff = current_time - session_data["created_at"]
if time_diff.total_seconds() > 3600: # 1 hour
sessions_to_remove.append(session_id)
for session_id in sessions_to_remove:
del user_sessions[session_id]
logger.info(f"Removed old session {session_id[:8]}...")
if sessions_to_remove:
logger.info(f"Cleaned up {len(sessions_to_remove)} old sessions")
class GradioUI:
"""A one-line interface to launch your agent in Gradio"""
def __init__(self, file_upload_folder: str | None = None):
self.file_upload_folder = file_upload_folder
if self.file_upload_folder is not None:
if not os.path.exists(file_upload_folder):
os.mkdir(file_upload_folder)
# Create reports directory
self.reports_folder = "reports"
if not os.path.exists(self.reports_folder):
os.mkdir(self.reports_folder)
def save_report(self, html_content: str) -> str:
"""Saves the HTML report and returns the file path."""
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
filename = f"vulnerability_report_{timestamp}.html"
filepath = os.path.join(self.reports_folder, filename)
with open(filepath, "w", encoding="utf-8") as f:
f.write(html_content)
return filepath
def validate_api_key(self, api_key: str) -> tuple[str, str]:
"""Validate API key and return status message."""
is_valid, message = validate_hf_api_key(api_key)
if is_valid:
return message, "success"
else:
return message, "error"
def interact_with_agent(self, prompt, messages, request: gr.Request):
"""Handle agent interaction with proper session management."""
# Get unique session ID for this user
session_id = get_user_session(request)
session_data = get_session_data(session_id)
logger.info(f"Processing request for session {session_id[:8]}...")
# Check if we have a valid agent for this session
if not session_data.get("agent"):
# Check if we have a valid HF_TOKEN in session
hf_token = session_data.get("hf_token")
# If no token in session, try to get it from .env file
if not hf_token:
env_token = os.getenv("HF_TOKEN")
if env_token:
hf_token = env_token
session_data["hf_token"] = env_token
session_data["max_steps"] = 10 # Default max_steps
logger.info(f"Using HF_TOKEN from .env file for session {session_id[:8]}...")
else:
logger.warning(f"No API key found for session {session_id[:8]}...")
error_msg = "❌ No API key provided. Please enter your Hugging Face API key in the API Configuration section above or set HF_TOKEN in your .env file."
messages.append(gr.ChatMessage(role="assistant", content=error_msg))
yield messages
return
logger.info(f"Creating agent for session {session_id[:8]}...")
if hf_token:
try:
max_steps = session_data.get("max_steps", 10)
session_data["agent"] = create_agent(hf_token, max_steps=max_steps)
logger.info(f"Agent created successfully for session {session_id[:8]}...")
except Exception as e:
logger.error(f"Failed to create agent for session {session_id[:8]}: {e}")
error_msg = f"❌ Failed to create agent with provided API key: {str(e)}"
messages.append(gr.ChatMessage(role="assistant", content=error_msg))
yield messages
return
else:
logger.info(f"Agent already exists for session {session_id[:8]}...")
# Adding monitoring
try:
# log the existence of agent memory
has_memory = hasattr(session_data["agent"], "memory")
print(f"Agent has memory: {has_memory}")
if has_memory:
print(f"Memory type: {type(session_data['agent'].memory)}")
# Prepare the system prompt
system_prompt = f"""You are a Vulnerability Intelligence Analyst. Complete the user request in {session_data.get('max_steps', 10)} steps maximum.
AVAILABLE TOOLS: nvd_search, web_search, cvedb_search, kevin_search, epss_search
CRITICAL RULES:
1. VERSION ANALYSIS:
- FIRST: Check current version via web search to understand the latest available version
- PRIORITY: When user asks for specific version, focus ONLY on vulnerabilities affecting that version and newer
- EXCLUDE OLDER: Do NOT report vulnerabilities that only affect older versions
- VERSION LOGIC:
* "up to X" or "before X" = affects versions UP TO X, NOT newer versions
* "X+" or "X and later" = affects X and newer versions
* "X through Y" = affects versions X to Y inclusive
- CORRECT LOGIC: If CVE affects "up to v22.1" and user asks about v24.0 then v24.0 is NOT vulnerable
- CORRECT LOGIC: If CVE affects "v25.0+" and user asks about v24.0 then v24.0 is NOT vulnerable
- CORRECT LOGIC: If CVE affects "below v25.0" and user asks about v24.0 then v24.0 is vulnerable
2. DATES: Use current date for "today" or "current". Use: from datetime import datetime; today = datetime.now().strftime("%Y-%m-%d")
3. PRODUCT SEARCH: ALWAYS use ONLY the base product name, NEVER include versions when using vulnerability tools (nvd_search, cvedb_search, kevin_search, epss_search)
4. SOURCES: Always prioritize vendor/original sources for CVE, CWE, and reference links (official vendor websites, security advisories)
5. SIMPLICITY: Keep code simple and logical. Avoid unnecessary library imports. Use only basic Python functions when needed.
6. TOOL USAGE: Use ONLY the available tools. Do not complicate tool calls with unnecessary code.
7. STRING OPERATIONS: Use simple Python methods (in, find, startswith, endswith, etc.). NEVER use .contains() - it doesn't exist in Python. Avoid complex string parsing of tool results.
8. VULNERABILITY ANALYSIS: When analyzing tool results:
- READ CAREFULLY: Pay attention to version ranges in vulnerability descriptions
- "up to X" means versions UP TO X, NOT including newer versions
- "below X" means versions BELOW X, NOT including X or newer
- "X+" means X and newer versions
- ONLY include vulnerabilities that actually affect the requested version
- If unsure about version compatibility, exclude the vulnerability
- DO NOT REASON: Don't create complex logic about version compatibility
- DO NOT ASSUME: If a CVE affects "up to 22.1", it does NOT affect 24.0
- SIMPLE RULE: Only include CVEs where the version range explicitly includes the requested version
9. REPORT GENERATION: Do NOT create complex functions, or loops for the final answer. Use the information collected and format it directly following the REPORT FORMAT.
REPORT FORMAT:
# Vulnerability Report
### [Software and Version]
#### CVE-ID: [CVE-YYYY-NNNNN]
**NIST NVD Link:** https://nvd.nist.gov/vuln/detail/CVE-YYYY-NNNNN
- **Attack Type:** [Type]
- **Published:** [Date]
- **CVSS:** [Score]
- **EPSS:** [Score]
- **KEV:** [Yes/No]
- **Affected Versions:** [Specific range]
- **Description:** [Description]
- **CWE:** [CWE-XXX] - https://cwe.mitre.org/data/definitions/XXX.html
- **Recommendations:** [Remediation advice]
- **Sources:**
- https://oficial-vendor-website.com/security-advisory
- https://official-security-source.com
- https://additional-reference-source.com
INSTRUCTIONS:
- Follow the exact format above
- Use official vendor sources when available
- Complete the task efficiently within the step limit
Now it is your turn, remember to keep code simple.
User Query: """
# Combine system prompt with user message
full_prompt = system_prompt + prompt
messages.append(gr.ChatMessage(role="user", content=prompt))
yield messages
logger.info(f"Starting agent interaction for session {session_id[:8]}...")
for msg in stream_to_gradio(
session_data["agent"], task=full_prompt, reset_agent_memory=False
):
# If the message contains an HTML report, we save it and update the message
if isinstance(msg.content, str) and msg.content.startswith("<!DOCTYPE html>"):
report_path = self.save_report(msg.content)
msg.content = f"Report generated and saved at: {report_path}\n\nYou can open the file in your browser to view the complete report."
messages.append(msg)
yield messages
# Clear sensitive data from session after interaction (AUTOMATIC)
clear_session_data(session_id)
logger.info(f"Session {session_id[:8]}... automatically cleared after interaction")
yield messages
except Exception as e:
logger.error(f"Error in interaction for session {session_id[:8]}: {str(e)}")
print(f"Error in interaction: {str(e)}")
error_msg = f"❌ Error during interaction: {str(e)}"
messages.append(gr.ChatMessage(role="assistant", content=error_msg))
yield messages
def setup_api_key(self, api_key: str, max_steps: int, request: gr.Request) -> str:
"""Setup API key for the user's session."""
# Get unique session ID for this user
session_id = get_user_session(request)
session_data = get_session_data(session_id)
logger.info(f"Setting up API key for session {session_id[:8]}...")
# Check if API key is provided from interface
if api_key and api_key.strip():
# Use the API key from interface
token_to_use = api_key.strip()
source = "interface"
else:
# Try to use token from .env file
env_token = os.getenv("HF_TOKEN")
if env_token:
token_to_use = env_token
source = ".env file"
else:
return "❌ No API key provided. Please enter your Hugging Face API key or set HF_TOKEN in your .env file."
# Validate the token
is_valid, message = validate_hf_api_key(token_to_use)
if is_valid:
# Store HF_TOKEN in session data
session_data["hf_token"] = token_to_use
session_data["max_steps"] = max_steps
logger.info(f"API key stored in session {session_id[:8]}... from {source}")
logger.info(f"Max steps set to: {max_steps}")
# Create new agent with the HF_TOKEN and max_steps
try:
session_data["agent"] = create_agent(token_to_use, max_steps=max_steps)
logger.info(f"Agent created successfully for session {session_id[:8]}...")
return f"✅ API key from {source} validated and agent created successfully! {message.split('!')[1] if '!' in message else ''}"
except Exception as e:
logger.error(f"Failed to create agent for session {session_id[:8]}: {e}")
return f"❌ Failed to create agent with API key from {source}: {str(e)}"
else:
logger.warning(f"Invalid API key for session {session_id[:8]}... from {source}")
return f"❌ Invalid API key from {source}: {message}"
def upload_file(
self,
file,
file_uploads_log,
allowed_file_types=[
"application/pdf",
"application/vnd.openxmlformats-officedocument.wordprocessingml.document",
"text/plain",
],
):
"""
Handle file uploads, default allowed types are .pdf, .docx, and .txt
"""
if file is None:
return gr.Textbox("No file uploaded", visible=True), file_uploads_log
try:
mime_type, _ = mimetypes.guess_type(file.name)
except Exception as e:
return gr.Textbox(f"Error: {e}", visible=True), file_uploads_log
if mime_type not in allowed_file_types:
return gr.Textbox("File type disallowed", visible=True), file_uploads_log
# Sanitize file name
original_name = os.path.basename(file.name)
sanitized_name = re.sub(
r"[^\w\-.]", "_", original_name
) # Replace any non-alphanumeric, non-dash, or non-dot characters with underscores
type_to_ext = {}
for ext, t in mimetypes.types_map.items():
if t not in type_to_ext:
type_to_ext[t] = ext
# Ensure the extension correlates to the mime type
sanitized_name = sanitized_name.split(".")[:-1]
sanitized_name.append("" + type_to_ext[mime_type])
sanitized_name = "".join(sanitized_name)
# Save the uploaded file to the specified folder
file_path = os.path.join(
self.file_upload_folder, os.path.basename(sanitized_name)
)
shutil.copy(file.name, file_path)
return gr.Textbox(
f"File uploaded: {file_path}", visible=True
), file_uploads_log + [file_path]
def log_user_message(self, text_input, file_uploads_log):
return (
text_input
+ (
f"\nYou have been provided with these files, which might be helpful or not: {file_uploads_log}"
if len(file_uploads_log) > 0
else ""
),
gr.Textbox(
value="",
interactive=False,
placeholder="Please wait while Steps are getting populated",
),
gr.Button(interactive=False),
)
def detect_device(self, request: gr.Request):
# Check whether the user device is a mobile or a computer
if not request:
return "Unknown device"
# Method 1: Check sec-ch-ua-mobile header
is_mobile_header = request.headers.get("sec-ch-ua-mobile")
if is_mobile_header:
return "Mobile" if "?1" in is_mobile_header else "Desktop"
# Method 2: Check user-agent string
user_agent = request.headers.get("user-agent", "").lower()
mobile_keywords = ["android", "iphone", "ipad", "mobile", "phone"]
if any(keyword in user_agent for keyword in mobile_keywords):
return "Mobile"
# Method 3: Check platform
platform = request.headers.get("sec-ch-ua-platform", "").lower()
if platform:
if platform in ['"android"', '"ios"']:
return "Mobile"
elif platform in ['"windows"', '"macos"', '"linux"']:
return "Desktop"
# Default case if no clear indicators
return "Desktop"
def launch(self, **kwargs):
with gr.Blocks(theme="ocean", fill_height=True) as demo:
# Different layouts for mobile and computer devices
@gr.render()
def layout(request: gr.Request):
device = self.detect_device(request)
print(f"device - {device}")
# Render layout with sidebar
if device == "Desktop":
with gr.Blocks(
fill_height=True,
):
file_uploads_log = gr.State([])
with gr.Sidebar():
# Project title and repository link at the top
gr.Markdown("""# Open Deep Research Vulnerability Intelligence""")
gr.Markdown("""<img src="https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png" width="20" height="20" style="display: inline-block; vertical-align: middle; margin-right: 8px;"> <a href="https://github.com/mcdaqc/open-deep-research-vulnerability-intelligence" target="_blank">Github Repository</a>""")
# About section
with gr.Accordion("ℹ️ About", open=False):
gr.Markdown("""**What it does:**
This AI agent specializes in automated vulnerability research and analysis, built on <a href="https://huggingface.co/blog/open-deep-research" target="_blank">Hugging Face's Open Deep Research</a> architecture. It can search across multiple security databases to provide comprehensive vulnerability intelligence reports.
**Available Tools & APIs:**
- <a href="https://nvd.nist.gov/developers/vulnerabilities" target="_blank">🛡️ NIST NVD</a> - National Vulnerability Database (free API)
- <a href="https://cvedb.com/" target="_blank">📊 Shodan CVEDB</a> - Comprehensive vulnerability database (free API)
- <a href="https://kevin.gtfkd.com/" target="_blank">⚠️ KEVin</a> - Known Exploited Vulnerabilities database (free API)
- <a href="https://www.first.org/epss/" target="_blank">📈 EPSS</a> - Exploit Prediction Scoring System (free API)
- 🌐 **Web Browser** - Navigate and extract information from web pages
**Model Configuration:**
- **Default Model**: Qwen/Qwen2.5-Coder-32B-Instruct (recommended)
- **Alternative**: You can also use Ollama with local models for privacy
**How to use:**
1. Enter your Hugging Face API key below
2. Ask about specific software versions, CVEs, or security vulnerabilities
3. The agent will automatically search all available databases
4. Receive comprehensive vulnerability reports with CVSS scores, EPSS predictions, and remediation advice""")
with gr.Group():
gr.Markdown("**Your request**", container=True)
text_input = gr.Textbox(
lines=3,
label="Your request",
container=False,
placeholder="Enter your prompt here and press Shift+Enter or press the button",
)
launch_research_btn = gr.Button(
"Run", variant="primary"
)
# Examples Section
with gr.Accordion("💡 Example Prompts", open=False):
gr.Markdown("**Click any example below to populate your request field:**")
example_btn_1 = gr.Button("🔍 MobaXterm 24.0 vulnerabilities", size="sm", variant="secondary")
example_btn_2 = gr.Button("🔍 Chrome 120.0.6099.109 security issues", size="sm", variant="secondary")
example_btn_3 = gr.Button("🔍 Apache Tomcat 9.0.65 KEV check", size="sm", variant="secondary")
example_btn_4 = gr.Button("🔍 Windows 11 recent vulnerabilities", size="sm", variant="secondary")
example_btn_5 = gr.Button("🔍 CVE-2024-0001 analysis", size="sm", variant="secondary")
example_btn_6 = gr.Button("🔍 Nginx 1.24.0 security status", size="sm", variant="secondary")
# Example button events
example_btn_1.click(
lambda: "Analyze MobaXterm 24.0 for vulnerabilities as of today",
None,
[text_input]
)
example_btn_2.click(
lambda: "Check Chrome 120.0.6099.109 for security vulnerabilities",
None,
[text_input]
)
example_btn_3.click(
lambda: "Is Apache Tomcat 9.0.65 in KEV database as of today?",
None,
[text_input]
)
example_btn_4.click(
lambda: "Check Windows 11 for recent vulnerabilities as of today",
None,
[text_input]
)
example_btn_5.click(
lambda: "Analyze CVE-2024-0001 in detail",
None,
[text_input]
)
example_btn_6.click(
lambda: "Check Nginx 1.24.0 for security vulnerabilities",
None,
[text_input]
)
# API Key Configuration Section
with gr.Accordion("🔑 API Configuration", open=False):
gr.Markdown("**Configure your Hugging Face API Key**")
gr.Markdown("🔒 **Security**: Each user has an independent session. API keys are automatically cleared after each interaction for your privacy.")
gr.Markdown("Get your API key from: https://huggingface.co/settings/tokens")
api_key_input = gr.Textbox(
label="Hugging Face API Key",
placeholder="hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
type="password",
lines=1
)
api_key_status = gr.Textbox(
label="Status",
value="✅ HF_TOKEN found in .env file. To use a different key, enter it above and click 'Setup API Key'." if os.getenv("HF_TOKEN") else "⚠️ Please enter your Hugging Face API key or set HF_TOKEN in your .env file.",
interactive=False
)
# Agent configuration
gr.Markdown("**Agent Configuration**")
max_steps_slider = gr.Slider(
minimum=5,
maximum=30,
value=10,
step=1,
label="Maximum Steps",
info="Number of steps the agent can take per session (higher = more detailed but slower)"
)
setup_api_btn = gr.Button("Setup API Key", variant="secondary")
# If an upload folder is provided, enable the upload feature
if self.file_upload_folder is not None:
upload_file = gr.File(label="Upload a file")
upload_status = gr.Textbox(
label="Upload Status",
interactive=False,
visible=False,
)
upload_file.change(
self.upload_file,
[upload_file, file_uploads_log],
[upload_status, file_uploads_log],
)
# Powered by smolagents
with gr.Row():
gr.HTML("""<div style="display: flex; align-items: center; gap: 8px; font-family: system-ui, -apple-system, sans-serif;">Powered by
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolagents/mascot_smol.png" style="width: 32px; height: 32px; object-fit: contain;" alt="logo">
<a target="_blank" href="https://github.com/huggingface/smolagents"><b>hf/smolagents</b></a>
</div>""")
# Chat interface
stored_messages = gr.State([])
chatbot = gr.Chatbot(
label="open-Deep-Research",
type="messages",
avatar_images=(
None,
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolagents/mascot_smol.png",
),
resizeable=False,
scale=1,
elem_id="my-chatbot",
)
# Add component to display reports
report_viewer = gr.HTML(label="Vulnerability Report", visible=False)
# API Key setup event
setup_api_btn.click(
self.setup_api_key,
[api_key_input, max_steps_slider],
[api_key_status]
)
text_input.submit(
self.log_user_message,
[text_input, file_uploads_log],
[stored_messages, text_input, launch_research_btn],
).then(
self.interact_with_agent,
[stored_messages, chatbot],
[chatbot],
).then(
lambda: (
gr.Textbox(
interactive=True,
placeholder="Enter your prompt here and press the button",
),
gr.Button(interactive=True),
),
None,
[text_input, launch_research_btn],
)
launch_research_btn.click(
self.log_user_message,
[text_input, file_uploads_log],
[stored_messages, text_input, launch_research_btn],
).then(
self.interact_with_agent,
[stored_messages, chatbot],
[chatbot],
).then(
lambda: (
gr.Textbox(
interactive=True,
placeholder="Enter your prompt here and press the button",
),
gr.Button(interactive=True),
),
None,
[text_input, launch_research_btn],
)
# Render simple layout
else:
with gr.Blocks(
fill_height=True,
):
# Project title and repository link at the top
gr.Markdown("""# Open Deep Research Vulnerability Intelligence""")
gr.Markdown("""<img src="https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png" width="20" height="20" style="display: inline-block; vertical-align: middle; margin-right: 8px;"> <a href="https://github.com/mcdaqc/open-deep-research-vulnerability-intelligence" target="_blank">Github Repository</a>""")
# About section for mobile
with gr.Accordion("ℹ️ About", open=False):
gr.Markdown("""**What it does:**
This AI agent specializes in automated vulnerability research and analysis, built on <a href="https://huggingface.co/blog/open-deep-research" target="_blank">Hugging Face's Open Deep Research</a> architecture. It can search across multiple security databases to provide comprehensive vulnerability intelligence reports.
**Available Tools & APIs:**
- <a href="https://nvd.nist.gov/developers/vulnerabilities" target="_blank">🛡️ NIST NVD</a> - National Vulnerability Database (free API)
- <a href="https://cvedb.com/" target="_blank">📊 Shodan CVEDB</a> - Comprehensive vulnerability database (free API)
- <a href="https://kevin.gtfkd.com/" target="_blank">⚠️ KEVin</a> - Known Exploited Vulnerabilities database (free API)
- <a href="https://www.first.org/epss/" target="_blank">📈 EPSS</a> - Exploit Prediction Scoring System (free API)
- 🌐 **Web Browser** - Navigate and extract information from web pages
**Model Configuration:**
- **Default Model**: Qwen/Qwen2.5-Coder-32B-Instruct (recommended)
- **Alternative**: You can also use Ollama with local models for privacy
**How to use:**
1. Enter your Hugging Face API key below
2. Ask about specific software versions, CVEs, or security vulnerabilities
3. The agent will automatically search all available databases
4. Receive comprehensive vulnerability reports with CVSS scores, EPSS predictions, and remediation advice""")
# API Key Configuration Section for Mobile
with gr.Accordion("🔑 API Configuration", open=False):
gr.Markdown("**Configure your Hugging Face API Key**")
gr.Markdown("🔒 **Security**: Each user has an independent session. API keys are automatically cleared after each interaction for your privacy.")
gr.Markdown("Get your API key from: https://huggingface.co/settings/tokens")
mobile_api_key_input = gr.Textbox(
label="Hugging Face API Key",
placeholder="hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
type="password",
lines=1
)
mobile_api_key_status = gr.Textbox(
label="Status",
value="✅ HF_TOKEN found in .env file. To use a different key, enter it above and click 'Setup API Key'." if os.getenv("HF_TOKEN") else "⚠️ Please enter your Hugging Face API key or set HF_TOKEN in your .env file.",
interactive=False
)
# Agent configuration for mobile
gr.Markdown("**Agent Configuration**")
mobile_max_steps_slider = gr.Slider(
minimum=5,
maximum=30,
value=10,
step=1,
label="Maximum Steps",
info="Number of steps the agent can take per session (higher = more detailed but slower)"
)
mobile_setup_api_btn = gr.Button("Setup API Key", variant="secondary")
# Chat interface for mobile
stored_messages = gr.State([])
file_uploads_log = gr.State([])
chatbot = gr.Chatbot(
label="open-Deep-Research",
type="messages",
avatar_images=(
None,
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolagents/mascot_smol.png",
),
resizeable=True,
scale=1,
)
# Mobile API Key setup event
mobile_setup_api_btn.click(
self.setup_api_key,
[mobile_api_key_input, mobile_max_steps_slider],
[mobile_api_key_status]
)
# Mobile Example button events
mobile_example_btn_1 = gr.Button("🔍 MobaXterm 24.0 vulnerabilities", size="sm", variant="secondary")
mobile_example_btn_2 = gr.Button("🔍 Chrome 120.0.6099.109 security analysis", size="sm", variant="secondary")
mobile_example_btn_3 = gr.Button("🔍 Apache Tomcat 9.0.65 KEV check", size="sm", variant="secondary")
# Mobile Example button events
mobile_example_btn_1.click(
lambda: "Analyze MobaXterm 24.0 for vulnerabilities as of today",
None,
[text_input]
)
mobile_example_btn_2.click(
lambda: "Check Chrome 120.0.6099.109 for current security issues",
None,
[text_input]
)
mobile_example_btn_3.click(
lambda: "Is Apache Tomcat 9.0.65 in KEV database as of today?",
None,
[text_input]
)
# Powered by smolagents for mobile
with gr.Row():
gr.HTML("""<div style="display: flex; align-items: center; gap: 8px; font-family: system-ui, -apple-system, sans-serif;">Powered by
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolagents/mascot_smol.png" style="width: 32px; height: 32px; object-fit: contain;" alt="logo">
<a target="_blank" href="https://github.com/huggingface/smolagents"><b>hf/smolagents</b></a>
</div>""")
text_input = gr.Textbox(
lines=1,
label="Your request",
placeholder="Enter your prompt here and press the button",
)
launch_research_btn = gr.Button(
"Run",
variant="primary",
)
# Examples Section for Mobile
with gr.Accordion("💡 Example Prompts", open=False):
gr.Markdown("**Click any example below to populate your request field:**")
mobile_example_btn_1 = gr.Button("🔍 MobaXterm 24.0 vulnerabilities", size="sm", variant="secondary")
mobile_example_btn_2 = gr.Button("🔍 Chrome 120.0.6099.109 security analysis", size="sm", variant="secondary")
mobile_example_btn_3 = gr.Button("🔍 Apache Tomcat 9.0.65 KEV check", size="sm", variant="secondary")
# Mobile Example button events
mobile_example_btn_1.click(
lambda: "Analyze MobaXterm 24.0 for vulnerabilities as of today",
None,
[text_input]
)
mobile_example_btn_2.click(
lambda: "Check Chrome 120.0.6099.109 for current security issues",
None,
[text_input]
)
mobile_example_btn_3.click(
lambda: "Is Apache Tomcat 9.0.65 in KEV database as of today?",
None,
[text_input]
)
# Powered by smolagents for mobile
with gr.Row():
gr.HTML("""<div style="display: flex; align-items: center; gap: 8px; font-family: system-ui, -apple-system, sans-serif;">Powered by
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolagents/mascot_smol.png" style="width: 32px; height: 32px; object-fit: contain;" alt="logo">
<a target="_blank" href="https://github.com/huggingface/smolagents"><b>hf/smolagents</b></a>
</div>""")
text_input.submit(
self.log_user_message,
[text_input, file_uploads_log],
[stored_messages, text_input, launch_research_btn],
).then(
self.interact_with_agent,
[stored_messages, chatbot],
[chatbot],
).then(
lambda: (
gr.Textbox(
interactive=True,
placeholder="Enter your prompt here and press the button",
),
gr.Button(interactive=True),
),
None,
[text_input, launch_research_btn],
)
launch_research_btn.click(
self.log_user_message,
[text_input, file_uploads_log],
[stored_messages, text_input, launch_research_btn],
).then(
self.interact_with_agent,
[stored_messages, chatbot],
[chatbot],
).then(
lambda: (
gr.Textbox(
interactive=True,
placeholder="Enter your prompt here and press the button",
),
gr.Button(interactive=True),
),
None,
[text_input, launch_research_btn],
)
# Start periodic cleanup of old sessions
def periodic_cleanup():
import time
while True:
try:
cleanup_old_sessions()
time.sleep(300) # Run every 5 minutes
except Exception as e:
logger.error(f"Error in periodic cleanup: {e}")
time.sleep(300)
cleanup_thread = threading.Thread(target=periodic_cleanup, daemon=True)
cleanup_thread.start()
demo.launch(debug=True, **kwargs)
# can this fix ctrl-c no response? no
try:
GradioUI().launch(mcp_server=True)
except KeyboardInterrupt:
...