Beta · Field intelligence powered by on-device AI
SiteGuard
Real-time AI monitoring for video and HLS streams. SiteGuard combines computer vision, language models, and rules engines to score incidents instantly, surfacing clear alerts on a dashboard your security team reads at 3 a.m.
— Operations Surface
Risk Score · Live Demo
Live feed mimics 24-hour site cameras. Scores update every 1.2s via Markov-chain generator.
0
Risk Score
Status: Nominal
What it does
- Multi-stream ingestRTSP, HLS, and file streams feed one unified analyzer.
- Risk scoringLangChain detectors score scene risk from 0 to 100 every second.
- Live dashboardsNext.js 15 and Recharts power chronological event review and deep incident drill-down.
- Auth + RBACSupabase auth with role-based access; multi-tenant by default.
- Alert pipelinesWebhook, email, and SMS sinks. Bring your own Twilio.
- GPU-ready DockerBuild CUDA-aware service profiles for compute stacks.
Tech
- FastAPI
- Next.js 15
- React 19
- LangChain
- OpenAI
- Supabase
- Docker Compose
- CUDA
Architecture
Three independent services ingest Worker converts inbound video into JPEG batches. analyzer Running detection models with LangChain for risk synthesis. web Tier serving a Next.js 15 dashboard on Supabase Postgres and storage. Each component scales independently. The compose file is the single source of truth, keeping local dev in sync with production.
Roadmap
- Ingest + dashboard MVPRTSP and HLS streaming pipelines, Supabase authentication, and a core dashboard.
- Detector library v2Swap detectors with versioned model manifests.
- SOC-2 readinessAudit trails, evidence gathering, and vendor oversight.
- On-prem applianceSingle app for air-gapped construction sites.