SentinelMesh

• 

  Main dashboard with 4 threat scenarios

• 

  Kill chain + incident + threat timeline

• 

  Ask ASI-1 copilot with suggested questions

• 

  Full dashboard with agents tab active

The Problem

Modern smart spaces face simultaneous cyber-physical attacks. A camera goes offline while unauthorized devices appear on the network. A badge is stolen while data exfiltration occurs. Today's tools treat these as separate problems.

The Innovation

SentinelMesh is the first multi-agent AI security copilot that fuses cyber and physical threat intelligence in real-time. Three specialist ASI-1 agents run in parallel:

• Cyber Analyst - Network forensics, C2 detection, exfiltration patterns
• Facility Safety - Physical access anomalies, environmental hazards
• Incident Commander - Cross-domain synthesis, executive briefing

How We Built It

We used Next.js 16 with TypeScript for a production-grade UI. The ASI-1 API is OpenAI-compatible, but it doesn't support native tool calling, so we built a custom workaround that pre-executes all 5 security tools and injects their results as context into the prompt.

We added interactive features: kill chain visualization mapped to MITRE ATT&CK, a what-if simulator to test containment strategies, and a live copilot that lets judges ask questions about incidents in natural language.

Key Challenges

ASI-1 kept returning 422 errors when we tried to pass the tools parameter. We solved this by pre-executing all tools upfront and injecting their results directly into the analysis prompt.

The UI was completely black at first because we were using transparent panels with near-identical background colors. Fixed by using solid panel backgrounds and better borders.

The scroll wasn't working in the bottom tabs panel. This was a Tailwind flex height chain issue that needed min-h-0 on the children.

We also added a 3-second instant demo mode so live judges don't have to wait 90 seconds for the API response.

What We Learned

ASI-1 requires creative API patterns. Native tool calling limitations actually pushed us toward smarter prompt engineering. Cyber-physical correlation is genuinely novel. And we learned that professional UI matters way more than backend polish at hackathons.

Built With

• asi-1-api
• next.js-16.2
• openai-sdk
• react-hooks
• recharts
• shadcn/ui
• tailwind-css-4
• typescript-5