Kryptos AI

Kryptos AI Logo
Kryptos Dashboard
Kryptos Extension popup

Inspiration

Kryptos AI was inspired by the rising complexity of digital deception. As scams move beyond simple text into "pixel-perfect" visual phishing and manipulative social engineering, traditional static blacklists are no longer enough. We wanted to build a "Vigilant Watchdog" that doesn't just block links but understands the intent behind a threat, protecting vulnerable populations from sophisticated scams and privacy violations.

What it does

Kryptos AI is a multi-agent Chrome extension that scans emails, messages, and web pages in real-time to prevent phishing and privacy violations. It features three core capabilities: real-time email and message scanning through text paste or screenshot OCR, an automated Privacy Policy analyzer that distills complex terms into plain English, and a predictive protection engine that warns users of trending scams before they interact with them.

How we built it

We built the system using a modular 4-agent architecture to allow simultaneous development. The Scout Agent monitors browser activity and handles screenshot OCR via Gemini Vision , while the Analyst Agent performs deep investigations by mapping threats to the MITRE ATT&CK framework. The Educator Agent then translates these findings into accessible voice and text alerts. The backend is orchestrated with FastAPI and uses Pydantic to maintain strict data contracts between agents.

Challenges we ran into

One of the primary challenges was managing the asynchronous handoffs between agents to ensure real-time performance without lagging the browser. Implementing the predictive protection system required complex MongoDB aggregations to identify trending threat patterns across the community while maintaining individual user privacy. We also had to overcome the strict lifecycle constraints of Chrome Extension Manifest V3 to maintain a persistent security state.

Accomplishments that we're proud of

We are particularly proud of our "Link Hover Intelligence," which provides safety ratings before a user clicks, and our ability to successfully map social engineering tactics directly to professional-grade threat classifications. Successfully integrating Gemini 2.0 Flash to handle sub-second reasoning across multiple agents was a major technical milestone that allowed our system to feel truly proactive.

What we learned

The weekend taught us the importance of enforcing strict API contracts in a collaborative AI environment to prevent integration bottlenecks. We learned how to utilize multimodal models like Gemini Vision to bridge the gap between visual deception and technical detection. Most importantly, we learned that effective security isn't just about blocking threats, but about creating transparency and awareness for the user.

What's next for Kryptos AI

We plan to expand Kryptos AI by deepening our predictive modeling to include demographic-specific threat alerts. We are also looking into expanding the "Watchdog" capabilities to mobile platforms to provide consistent protection across all user devices. Future iterations will include even more granular privacy controls, allowing users to automatically opt-out of data collection detected by our Privacy Policy analyzer.