Reflects actual fleet state: 44 Ollama models, 173 TCP ports, 5/5 nodes, stats API, analytics system, all websites pulling live data. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> RoadChain-SHA2048: fa7bff19fa8dbb3b RoadChain-Identity: alexa@sovereign RoadChain-Full: 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
4.9 KiB
BlackRoad OS
Self-hosted edge AI operating system running 52 TOPS of inference across a 5-node Raspberry Pi cluster.
What is BlackRoad OS?
BlackRoad OS is a sovereign computing platform that runs AI inference, networking, and automation entirely on hardware you own. No cloud dependencies. No API keys expiring at 2am. Five Raspberry Pi 5 nodes connected by a WireGuard mesh network, accelerated by dual Hailo-8 NPUs, running 44 local models through Ollama.
The fleet self-heals. Cron-driven watchdogs monitor every service, restart what dies, and log what broke. A stats API collects live telemetry from every node every 5 minutes — CPU temp, RAM, disk, services, model counts — and serves it to all BlackRoad websites in real time. Cloudflare tunnels route 48+ domains through the cluster. Docker Swarm orchestrates containers across nodes. Gitea hosts 207 repositories on a 1TB NVMe.
This is infrastructure that runs in a closet and serves production traffic.
Live Fleet Status
Real-time data from the stats API (updated every 5 minutes):
| Node | Role | Hardware | Key Services |
|---|---|---|---|
| Alice | Gateway, DNS, ingress | Pi 400 | Pi-hole, PostgreSQL, Qdrant, Nginx, 65+ tunnel routes |
| Cecilia | AI inference, TTS | Pi 5, Hailo-8 (26 TOPS) | 15 Ollama models, CECE API, TTS, MinIO, PostgreSQL |
| Octavia | Git, orchestration, AI | Pi 5, Hailo-8 (26 TOPS), 1TB NVMe | 11 Ollama models, Gitea, Docker Swarm, NATS |
| Aria | Container management | Pi 5 | 6 Ollama models, Portainer, Headscale |
| Lucidia | Apps, CI/CD | Pi 5, 238GB SD | 6 Ollama models, GitHub Actions, FastAPI, CarPool, PowerDNS |
Network: WireGuard mesh (10.8.0.x) with Anastasia (DigitalOcean NYC) as hub. RoadNet WiFi overlay (5 APs, dedicated subnets).
By the Numbers
| Metric | Count |
|---|---|
| Edge nodes | 5 Raspberry Pis |
| AI compute | 52 TOPS (2x Hailo-8) |
| Ollama models | 44 across fleet |
| TCP listening ports | 173 |
| Cloudflare tunnels | 18 (100 hostnames) |
| Custom domains | 48+ |
| Pages sites | 95 |
| D1 databases | 8 |
| KV namespaces | 40 |
| R2 buckets | 10 |
| Gitea repos | 207 |
| GitHub repos | 115 (non-fork) |
| SQLite databases | 228 |
| Shell scripts | 92 |
| Cron jobs | 13 automated tasks |
Stack
| Component | What it does |
|---|---|
| Hailo-8 | 52 TOPS combined neural network inference (2x 26 TOPS) |
| Ollama | 44 local LLMs including custom CECE models |
| WireGuard | Encrypted mesh VPN across all nodes + cloud |
| Pi-hole | Fleet-wide DNS filtering (76,456 blocked domains) |
| Docker Swarm | Container orchestration across the cluster |
| Gitea | Self-hosted Git with 207 repositories across 7 orgs |
| NATS | Lightweight messaging between services |
| RoadC | Custom programming language with Python-style indentation |
| Cloudflare | 18 tunnels, 95 Pages, D1/KV/R2 storage, Workers |
| Stats API | Live fleet telemetry via Cloudflare Worker + KV |
| Analytics | Sovereign, cookie-free event capture via Worker + D1 |
Observability
Every BlackRoad website pulls live data from the stats API — no hardcoded numbers. The collector script SSHes into all 5 nodes every 5 minutes and pushes:
- CPU temperature, usage, RAM, disk
- Ollama model counts and service status
- Docker container counts and TCP port inventory
- Service health (SSH, DNS, Ollama, Nginx, PostgreSQL, MinIO, Gitea, NATS, etc.)
Websites auto-refresh every 60 seconds. When a node goes down, every dashboard reflects it within 5 minutes.
Sites
| Site | Purpose |
|---|---|
| blackroad.io | Main platform |
| blackroad.network | Live mesh topology dashboard |
| blackroad.systems | Infrastructure status page |
| backlog.blackroad.io | Engineering journal (35 articles) |
| git.blackroad.io | Self-hosted Gitea |
License
Copyright 2026 BlackRoad OS, Inc. — Alexa Amundson. All rights reserved.
Related Projects
| Project | Description |
|---|---|
| Lucidia | Autonomous AI agent with persistent memory |
| RoadC | Custom programming language with indentation-based syntax |
| Quantum Math Lab | Mathematical proofs and quantum computing research |
| Hailo Vision | Computer vision with Hailo-8 AI accelerators |
| Fleet Heartbeat | Distributed health monitoring system |
| BlackRoad API SDKs | Python, Go, TypeScript SDKs for the BlackRoad API |
| BlackRoad KPIs | Daily fleet metrics collection and reporting |