agentic-chatops

Production agentic ChatOps platform implementing all 21 design patterns from [Agentic Design Patterns](https://drive.google.com/file/d/1-5ho2aSZ-z0FcW8W_jMUoFSQ5hTKvJ43/view?usp=drivesdk) by Antonio Gulli — running on a self-hosted homelab managed by a single operator.

---

Managing 310 infrastructure objects — 113 physical devices, 197 virtual machines, 421 IP addresses, 39 VLANs, 653 interfaces across 6 sites (Netherlands, Greece x2, Switzerland, Norway) and 3 Proxmox clusters — as a solo operator is unsustainable without automation.

That's 3 firewalls, 3 managed switches, 12 Kubernetes nodes with Cilium ClusterMesh, self-hosted everything (Matrix, GitLab, YouTrack, n8n, LibreNMS, Grafana, Nextcloud HA, SeaweedFS, Thanos), and no team to delegate to. When an alert fires at 3am, there's one person on call. Always.

This platform bridges the gap: infrastructure alerts flow in, AI agents triage and investigate, propose remediation plans, and wait for human approval before executing. The human stays in the loop for critical decisions but doesn't have to do the detective work.

LibreNMS/Prometheus Alert
         │
         ▼
   ┌─────────────┐     ┌──────────────┐     ┌──────────────┐
   │  n8n         │────▶│  OpenClaw     │────▶│  Claude Code  │
   │  Orchestrator│     │  (Tier 1)    │     │  (Tier 2)     │
   │  10 workflows│     │  GPT-4o      │     │  Claude Opus  │
   │  ~340 nodes  │     │  L1/L2 triage│     │  Deep analysis│
   └──────┬───────┘     └──────────────┘     └───────┬───────┘
          │                                          │
          ▼                                          ▼
   ┌─────────────┐                            ┌──────────────┐
   │  Matrix      │◀───────────────────────────│  Human (T3)  │
   │  Chat rooms  │  polls, reactions, replies │  Approval    │
   └─────────────┘                            └──────