I love self-hosting. I run my own infrastructure, keep my data on systems I control, and host as much of my software as I reasonably can, including AI models.

The fun for me is owning the whole stack: hardware, virtualization, storage, networking, containers, monitoring, backups, applications, and the automation that keeps it running. I like being able to trace a problem from a UI all the way down to a disk or network path and fix it myself.

My home infrastructure has a six-host Proxmox cluster and a Docker Swarm with six active nodes. The Swarm currently runs more than 100 services across more than 50 stacks. Separate storage and backup systems handle bulk data, snapshots, and replicas.

I also run dedicated GPU servers for local language models, speech, and image generation. Around that, I self-host source control, CI/CD, a container registry, monitoring and alerting, home automation, media, personal applications, and the AI-agent environment I use every day.

Ceph and CephFS provide shared distributed storage across the virtualization and container layers. ZFS handles host-local application data, snapshots, and replication. I use Ansible to keep host and guest configuration repeatable instead of relying on a collection of one-off manual setups.

I built it because I want control over my own data and systems. I can understand or change every layer and keep useful software running without depending on someone else's cloud.

AI agents are part of how I get more out of the infrastructure. They have enabled me to take on more projects and automate more of the homelab's routine management. My setup includes a downstream Hermes Agent distribution, plugins, MCP servers, persistent memory, RAG over my own documents and operational knowledge, scheduled workflows, delegated agents, mobile and web access, voice, and routing across local and cloud models.

Agents can monitor services, investigate issues, coordinate tasks, and carry routine work forward autonomously. Consequential infrastructure changes still require explicit human approval.

I also like experimenting with AI software factories and agent-graph setups.

Dedicated GPU servers let me run language models and generative workloads locally. That gives me a private option for prompts and data I do not want to send elsewhere, while still letting me use cloud models when they are the better tool.

I am most interested in making agents useful over time: durable sessions, clear tool boundaries, human approval where it matters, observable behavior, and recovery when something fails.

• Personal-finance software and paper-trading research systems
• Offline maps, geospatial pipelines, and self-hosted news intelligence
• Infrastructure automation, release tooling, monitoring, backup, and recovery workflows

Most of this work is private or runs only on my home infrastructure.