I started working on AGENT AEGIS after thinking about a simple problem: if AI agents can make decisions and take actions on their own, who makes sure they don't go too far?
That became the idea behind Aegis — a safety layer that checks an agent's actions before they reach the real world. It can evaluate risk, check policies, allow safe actions, ask for human approval, or block dangerous actions. I also added threat monitoring, an audit trail, and a scenario builder so the whole process can be explored interactively.
I built the project with Claude, HTML, CSS, and JavaScript, and deployed it using Vercel. The hardest part was turning a fairly complex AI-safety idea into something that people could understand and interact with immediately.
The goal: make AI more capable without making humans lose control.
Built With
- anthropic
- claude
- css
- html
- javascript
- vercel
