A self-hosted homelab running 40+ Docker containers. It carries family photo and video albums, a media library streamed from my own DVD collection, my personal portfolio, and privacy-first replacements for tools I used to rent. I manage the whole stack declaratively in Komodo and operate it largely by intent: I describe the outcome I need and an AI/MCP control plane spins up containers, investigates resource and log issues, and tunes the stack. That operating model is what finally made self-hosting worth my time, and it lets me trust the stack with data I can't afford to lose.
Agent-Operated Container Stack
Self-hosting / Infrastructure / Cloud-native opsUndated
Role: Docker wrangler & AI whisperer
40+
Docker containers
Run from a single declarative Komodo control plane.
1–2 yrs
Continuously operated
A live stack my family relies on daily.
Approximate
3-2-1
Off-site backup
Scheduled backups with Backblaze as the off-site copy and a proven restore after a bad change.
By intent
AI / MCP operations
I describe the outcome; an agent control plane deploys, diagnoses, and tunes.
Overview
A homelab earns a place in a product portfolio because it is where I operate real infrastructure from the owner's seat. It runs 40+ Docker containers managed declaratively in Komodo, carrying things my family actually depends on: photo and video albums, a media library streamed from my own DVD collection, my personal portfolio site, and privacy-first replacements for tools I used to rent.
Operating the stack by intent
I avoided serious self-hosting for years because the upkeep cost more time than it saved. Wiring the stack to an AI/MCP control plane, off-the-shelf and custom, changed the economics. I now manage containers at the level of intent rather than mechanics: I describe the solution I need and it gets deployed, I ask it to find and correct what is consuming resources, and I have it investigate application logs and propose fixes. Stack management, not babysitting.
Data I can't afford to lose
Once family photos and videos live on hardware I own, backup stops being theoretical. Everything important is protected with scheduled backups on a 3-2-1, off-site pattern, with Backblaze holding the off-site copy. I have recovered for real. After a bad change corrupted state, I restored from a known-good backup instead of losing data. Tested recovery, not just configured backup.
The tradeoff I keep making
Self-host versus rent. I bring workloads in-house when control, privacy, and recurring cost beat the convenience of SaaS, down to building web apps with Claude Code and Codex and my own technical skill instead of paying for AI-as-a-service builders. Every container on the stack is a small, deliberate decision about whether the operating cost is worth the value it returns.
- Family photo and video albums
- Personal media streaming from my own DVD collection, with automated library management
- My own network core: router, DHCP, and DNS
- Web hosting for my portfolio site and personal web apps
- Privacy-first self-hosted replacements for rented SaaS
What runs on it
- Family photo and video albums
- Personal media streaming from my own DVD collection, with automated library management
- My own network core: router, DHCP, and DNS
- Web hosting for my portfolio site and personal web apps
- Privacy-first self-hosted replacements for rented SaaS
How I operate it
- Declarative stack management in Komodo, configuration as code
- Deploy new containers by describing the outcome, not the steps
- Agent-driven diagnosis of resource consumption and application logs
- Off-the-shelf and custom MCP integrations as the control plane
Data protection
- Scheduled backups via Duplicati / Backrest
- 3-2-1 pattern with Backblaze as the off-site copy
- Recovered for real: restored from a known-good backup after a bad change corrupted state
- Tested recovery, not just configured backup
Decisions and tradeoffs
- Self-host versus rent, decided per container on cost, privacy, and control
- Brought web-app building in-house with Claude Code and Codex instead of AI-as-a-service
- Accept operational burden where owning the data is worth it
Want a PM who operates the stack, not just the roadmap?
I form product convictions from hands-on operation: deploying, diagnosing, and recovering real systems myself.