h-srv — Services & Orchestration
Central server: HAProxy reverse proxy, Grafana + TimescaleDB metrics, NetBox (source of truth), h-cli AI engine, FRR (BGP/OSPF).
This lab runs the same patterns I recommend to clients: proper BGP hub-spoke design, centralized monitoring with time-series metrics, and network automation driven by a source of truth. No shortcuts — production-grade infrastructure at a smaller scale.
Central server: HAProxy reverse proxy, Grafana + TimescaleDB metrics, NetBox (source of truth), h-cli AI engine, FRR (BGP/OSPF).
Dual RTX 5090 (32GB each). Local LLM inference. Builds every vector database that powers h-cli's vendor knowledge.
EVE-NG Pro, unlimited topologies. RTX 3090 + RTX 4070 Ti for offline h-cli inference. CLOS / EVPN / MPLS fabric testing.
── Lab Topology: CLOS Fabric, gateway at spine (Exit-at-Spine) ──
┌──────┐ ┌──────┐
│ SP-1 │ │ SP-2 │──eBGP──► physical network (AS 65000)
└──┬┬──┘ └──┬┬──┘
┌────┘└────┐┌───┘└────┐
┌────┴──┐ ┌────┴┴──┐ ┌───┴───┐
│ LF-1 │ │ LF-2 │ │ LF-3 │
└───┬───┘ └───┬────┘ └───┬───┘
│ │ │
hosts hosts hosts This is a homelab in name only — proper BGP hub-spoke design, source-of-truth-driven automation, and full observability, at a scale you can iterate on before it hits production.