ForgeScaler

The secure OS powering VaultScaler's edge AI observability and orchestration systems.

The Nervous System of VaultScaler

ForgeScaler is a memory-driven, DevOps-inspired OS purpose-built for edge AI nodes like VaultScaler One (Jetson Orin NX) and VaultScaler Micro (Raspberry Pi 5). It orchestrates real-time facial recognition, threat detection, and intelligent alerting — all offline-first, with secure OTA update pipelines and full observability.

Unlike cloud inference systems, ForgeScaler runs fully on-prem. Logs are journaled to `.jsonl`, updates are cryptographically signed, and model execution remains sovereign and local.

System Architecture

⚙️ Core Platform

  • k3s + Helm on Jetson Orin / Pi 5
  • OSTree-style immutable images
  • OTA via GitHub Actions
  • SQLite + S3 sync for logs

🤖 Edge Agents

  • bootstrap-agent — secure first boot
  • security-inference-agent — model runner
  • reflector-agent — logs `.jsonl` events
  • ota-agent — signed OTA validation
  • triage-agent — local failure diagnosis

🧠 Memory System

  • Append-only `.jsonl` logfiles
  • Markdown-based retrospectives
  • Local dashboard view (optional)

🔐 Security & Privacy

  • Zero cloud dependency
  • Zero-trust LAN API
  • Facial database lives locally
  • No vendor lock-in — containerized

Hardware Targets

Why ForgeScaler Exists

Most AI systems require the cloud to function — leaking metadata, enabling surveillance, and breaking down during outages. ForgeScaler rethinks this stack: a self-contained orchestration layer that works without the internet, respects privacy, and enables true operational ownership.

This is not just DevOps. This is RecursiveOps — infrastructure that reflects on itself, version-controls its memories, and evolves alongside its operators.

View on GitHub