Author
154 articles
Milan Buha is the founder and editor of VoidInsider. Curious by default, he likes understanding things from the inside — the technology he depends on every day, the systems behind everyday money decisions, the fine print everyone else skips. He writes about what he's tried, checked, and actually uses, the way he'd explain it to a friend.
Every self-hosted apps roundup is an unvetted list with no honest 'skip this one.' Six categories actually tested on a homelab — AI chatbot, music, automation, Reddit, Git, and CI/CD — with real hardware cost and one category that isn't worth it for most people.
Price and sound quality rarely keep people on Spotify -- years of playlists do. A real migration run with two transfer tools, tracking exactly what transferred, what needed manual fixing, and the real free-tier limits on each.
Gitea and Forgejo run on well under 1GB of RAM; GitLab CE documents a 4GB minimum. This comparison covers real resource footprint, CI runner maturity (Gitea/Forgejo Actions vs GitLab CI), and GitHub migration friction for homelab hardware.
Redlib is the actively maintained fork of the archived Libreddit -- a self-hosted, ad-free, tracker-free Reddit frontend. It's easy to run in Docker, but Reddit actively rate-limits and blocks it, so public instances go down often and self-hosting your own copy only shifts who owns that problem.
A Raspberry Pi makes a genuinely good local-only security camera — if you get the two decisions right: which AI accelerator (the Coral is now deprecated; the Hailo-8L AI HAT+ is the pick) and how much storage continuous recording really eats. Real per-camera limits, worked storage math, and a Frigate config that runs.
A rack of Raspberry Pis looks great, but is a cluster the right box? The honest version: the PoE power-budget math nobody calculates, real Pi 5 thermals in a dense rack, and a straight cluster-versus-single-mini-PC verdict. Most people who want a cluster actually need a mini PC.
A spare Pi and a dozen contradictory tutorials -- here's the map. Real tested ceilings for the six things people actually self-host on a Raspberry Pi: NAS throughput, Minecraft player caps, camera storage math, retro-gaming playability, cluster power, and remote access, plus where to reach for a mini PC instead.
A real config.sh/svc.sh walkthrough for a self-hosted GitHub Actions runner, an honest spec comparison against GitHub-hosted minutes, and why GitHub itself warns against running one on a public repo (fork PRs can execute code on it).
A ChatGPT price hike or a moment of regret pasting something sensitive into a public chatbot raises the obvious question: can your homelab box just run the AI itself? Real tokens-per-second numbers across Raspberry Pi 5, N100 mini PC, and GPU tiers -- plus what a local LLM can and can't replace.
A weekend spent importing a ripped CD or FLAC collection raises the same question: which self-hosted server to stream it back with. Real resource footprints for Navidrome, Airsonic, and Jellyfin, plus a mobile-app compatibility matrix and why Airsonic isn't a safe pick for a new build in 2026.
n8n Cloud stops every workflow the moment you hit the monthly execution cap. What self-hosting the free Community edition actually gets you and gives up -- real Docker resource usage, the Enterprise-only feature gaps, and the dollar math against Cloud's Starter plan.
How many players can a Pi really hold? Realistic PaperMC ceilings for Pi 4 vs Pi 5, the Bedrock ARM problem, the tuning levers, and a Docker setup.