Platform · Live · always-on · 2026 – Present
AIOS Self-hosted AI operations system.
Architect · builder · operator
A private, always-on operations backbone: a local model fleet, a NATS JetStream event bus, a routing layer that dispatches each request to the right execution engine, and a unified Mission Control dashboard for service health, live jobs and scheduled automations.
Designed and built a private, always-on operations backbone that lets autonomous agents research, read and write files, and produce documents unattended, engineered to operations standard with failover, watchdogs and backups.
- Tool servers wired
- 7 MCP
- Knowledge base
- 40k+ chunks
- Ollama
- NATS JetStream
- MCP
- Qdrant
- n8n
- Docker
- Tailscale
- Cloudflare Tunnel
What I built.
Event-driven core.
A local model fleet, a NATS JetStream event bus, and a routing layer that classifies each request and dispatches it to the right execution engine.
Connector layer: 7 MCP tool servers.
Filesystem, knowledge base, web research, document generation, plus a self-service panel with token and browser-OAuth onboarding, so agents work unattended.
Autonomous scheduled agents.
Including a daily research brief that browses live sources, compiles findings and delivers them to a messaging channel without supervision.
Cited retrieval.
A retrieval index over a 40,000+ chunk private knowledge base for cited internal search.
Reliability layer.
Primary/backup failover, a 60-second watchdog with circuit breaker and alerting, auto-restart of every service on reboot, and daily configuration backups (47 critical files) with 30-day retention and off-drive mirroring.
What changed.
- MCP tool servers
- 7
- Knowledge-base chunks indexed
- 40,000+
- Watchdog cycle
- 60s
- Critical files backed up daily
- 47