Cyrus - Sentinel Security

Modern cybersecurity suffers from a critical timing problem: zero-day vulnerabilities and complex logic flaws are often discovered too late—during final audits or, worse, in production. The idea for Cyrus was born from an absolute necessity: to bring the power of a reverse engineering and pentesting expert directly into the developer's IDE (VS Code), in real time.

⚙️ What it does Cyrus is not a simple autocomplete chatbot. It is an autonomous security agent integrated into VS Code. While the developer codes, the agent:

Analyzes the global context: It reads the project architecture in the background.

Hunts for vulnerabilities: It detects logic flaws, injection risks, and poor security practices.

Proposes fixes: It pedagogically explains the flaw line by line and generates a secure patch.

🛠️ How we built it Cyrus relies on a modern architecture designed for speed and depth of analysis:

The Brain (Gemini 1.5): We leverage the Google Gemini 1.5 Flash/Pro API. Thanks to its massive context window, the agent can understand the entire code repository (multiple files) to find complex flaws that require a global architectural understanding.

Agentic Capabilities (Function Calling): The agent uses Gemini's Function Calling to autonomously decide to open and read other related files (like a database or config file) if it suspects a cascading flaw.

Backend: A Node.js server handles the agent's logic and communication with the Google API.

Frontend: A native VS Code extension built in TypeScript for a seamless and non-intrusive UI.

🚧 Challenges we ran into Transforming an LLM into a truly autonomous agent (rather than a simple interactive assistant) required rigorous prompt engineering. We had to calibrate the "System Prompt" to force Gemini into a strict cybersecurity auditor mindset, returning its results in a structured JSON format for the extension, all while preventing hallucinations.

🏆 Accomplishments that we're proud of Successfully designing a workflow where security becomes proactive rather than reactive. Cyrus proves that with Gemini's large context window, AI can understand a developer's architectural intent and act as an invisible, intelligent safety net.

🚀 What's next for Cyrus Our goal is to expand Cyrus's agentic capabilities. Next steps include:

Integrating automated penetration testing via disposable virtual terminals.

Real-time network traffic analysis from the development environment.

Evolving Cyrus into a complete, standalone security OS tailored for pentesters.

Built With

• google-gemini
• node.js
• typescript
• vscode-extension