burntai.com — AI Infrastructure Engineering

// about

This is the field journal of ClydeStack — a self-hosted AI infrastructure built from commodity hardware, open-source tooling, and a lot of late nights.

Not a tutorial site. Not a product. A living system being engineered in the open.

// live status

Hosts Online

—

loading…

Active Agents

—

loading…

Projects

—

loading…

Platform

Clyde
Stack

Proxmox · Ceph · Vault

// the stack

L1

Hardware

3× GMKtec NucBox K7+ (Proxmox cluster) · Beelink SER (DR) · OPNsense N100 gateway · Ubiquiti switch · UPS

L2

Virtualisation

Proxmox VE 3-node cluster · Ceph distributed storage (~900 GiB usable) · LXC containers · VLAN segmentation (T1/T2/T3/DMZ/IoT)

L3

Platform

HashiCorp Vault OTP SSH · PowerDNS primary+secondary · PostgreSQL 16 + hot standby · Prometheus + Grafana · rsyslog

L4

AI Orchestration

ClydeNexus command center · AgentDeck (cc_agent fleet, 14 hosts) · ProjectLens · Anthropic API · Ollama qwen2.5:32b on RX 6900 XT

L5

Applications

PAI (personal AI server, 4 users) · Sentinel (threat analysis, 14 hosts monitored) · jobs.burntai.com · ClydeLog Analyst

// build log

2026-04-25

securityIoT VLAN segmentation — 10 devices isolated, Bambu P1S + Alienware explicit pass-through

2026-04-23

aiqwen2.5:32b deployed to clyde — fleet LLM upgraded from 8B to 32B, ROCm-accelerated on RX 6900 XT

2026-04-22

resiliencePostgreSQL streaming replication live — db2 hot standby, 0 lag, auto-failover via DNS flip

2026-04-21

infraclyde: Windows wiped → Ubuntu 24.04, ROCm 6.3 native, all 6 fleet services LIVE

2026-04-20

infraFull Proxmox migration complete — every VM/LXC moved off Hyper-V to pve1/pve2/pve3

2026-04-17

appVLAN segmentation + dmzproxy live — jobs.burntai.com launched in DMZ with TLS + Google OAuth

2026-04-13

infraPowerDNS primary+secondary LIVE — fleet-wide .clydestack DNS with PTR zones

2026-04-08

aiClydeLog Analyst — Haiku proactive log review, 6-hour sweep + daily digest via Telegram

2026-04-07

infra3-node Proxmox cluster + Ceph LIVE — 927 GiB usable, all LXCs migrated to distributed storage

2026-04-04

resilienceOPNsense fw1 + MiFi WAN2 auto-failover — ~15s WAN failover tested and confirmed

2026-03-30

aiTemporal awareness — gap detection on startup, multi-source gap reports, Telegram bot LIVE

2026-03-21

securityVault SSH OTP fleet-wide — cladmin + vault-ssh-helper on all hosts, break-glass in Vault KV

// links

→ jobs.burntai.com AI-assisted job tracking · Gmail smart scan · Claude Haiku extraction

→ about Jeff Perry · linkedin · email