TL;DR – I plan my side project with a coding agent in OpenProject instead of a chat window. I write user stories (what and why); the agent, working as its own least-privilege user, turns them into features and tasks with specs (how) and hands them back for review. It only writes code for items assigned to it, and only when I ask. It works, it costs some ceremony, and the MCP tooling still has rough edges.

My homelab is mostly text: Docker Swarm compose files, bootc Containerfiles, Forgejo workflows, a few Ansible playbooks and this Hugo blog. All of it lives in git, and for years the workflow has been the same: edit in an editor, render, deploy, check the logs, repeat.

Over the last months two coding agents moved into that loop: opencode and Claude Code. This post covers how I set them up, what they do for me, and where I still keep my hands on the wheel.

Every prompt you send to a hosted LLM leaves your network. That includes the text you type, the files you attach, and whatever context your tools decide to pack around it. For a quick question about a regex that doesn’t matter. For your mailbox, your customers’ tickets or your company wiki, it matters a lot.

That is the main reason I run most of my models on my own hardware with Ollama. This post covers why, and where these models are useful in a self-hosted setup.