Neural-Trace is a sophisticated security auditing agent built to bridge the gap between low-level x86 Assembly instructions and high-level security intelligence. By leveraging the advanced reasoning of Gemini 3, it automates the detection of critical vulnerabilities in legacy binaries.
- Deep-Think Logic Simulation: Utilizes Gemini 3's
thinking_level="HIGH"to simulate CPU register states and memory stack behavior, exposing the model's internal "thought process". - Agentic Vulnerability Mapping: Detects buffer overflows, integer overflows, and logic bombs with high precision.
- Verified Exploitation: Uses Gemini's built-in Python Code Execution tool to mathematically verify memory displacements and potential overflow offsets.
- Remediation Engine: Automatically generates memory-safe Python rewrites of vulnerable logic to assist in modernizing legacy systems.
- Python & Streamlit: For the orchestration engine and interactive forensic dashboard.
- Google Gemini 3 API (Flash): The core reasoning and code execution engine.
- x86 Assembly: The target forensic environment.
- Ingestion: Upload a raw
.asmor.txtbinary dump. - Audit: The agent performs a forensic scan, identifying insecure memory management.
- Verification: Using Native Code Execution, the agent calculates the exact byte-offsets required for privilege escalation.
- Reporting: A full forensic report is generated, including the AI's step-by-step reasoning chain.
- Clone the repo:
git clone https://github.com/Sutharshannn/Neural-Trace.git - Install dependencies:
pip install streamlit google-genai python-dotenv - Add your
GEMINI_API_KEYto a.envfile. - Run the app:
streamlit run app.py
Developed as a capstone-level project during the Gemini 3 Hackathon (Feb 2026). Author: Sutharshan Suthakaran
Disclaimer: This tool is intended for ethical security research and educational purposes only.