Setup, power and thermals, and software tips for running a Mac mini as a home server or self-hosting box.
Vaultwarden was running on an internal server behind nginx proxy manager with a Let’s Encrypt certificate, and the setup worked for more than a year. A few days ago, access stopped from every device and browser. Running curl against the address returns a TLS connection error that appears to involve SNI checking. The current nginx proxy manager setup runs on the host network and forwards HTTP traffic to 127.0.0.1 on port 18000. The SSL settings have Force SSL, HTTP/2, HSTS, and HSTS subdomains enabled. Common exploit blocking and Websocket support are also enabled, and the advanced proxy settings include disabled buffering, no request body size limit, HTTP 1.1, and upgrade header forwarding.
Self-hosting Obsidian sync keeps personal notes under personal control, but it also turns note syncing into another system to maintain. Git-based syncing and CouchDB/LiveSync experiments can work, yet a small markdown sync problem can consume an evening of troubleshooting. Easy services such as Obsidian Sync or Dropbox rely on US cloud hosting, which is a problem for someone who wants notes to stay in the EU. Private options such as running LiveSync on a VPS or NAS avoid some cloud concerns, but they still need patches, monitoring, and care. The practical choice becomes a tradeoff between full self-hosting and paying for a managed service that is EU-hosted, encrypted, and keeps data ownership clear.
A university OneDrive account offers up to 10TB of storage, but it is hard to use smoothly from an Ubuntu 26.04 work laptop. The built-in OneDrive connection can show files, but it does not work well for editing them as part of normal file work. The laptop has only 250GB of storage, so it cannot hold all the large data from the research project locally. rclone sync and an open-source OneDrive tool were both tried, but the university would not allow either tool to access the account. The university IT office refused approval and only allowed manual uploads and downloads through the web interface. That manual process is not practical for ongoing work with large data.
Several services need to be reachable from the public internet, but without requiring a VPN. Tailscale worked before, but it was hard for less technical people to use. Some services currently run on a VPS with Caddy acting as a reverse proxy, while others run inside a homelab. The main choice is whether to move to Pangolin or add an authentication tool such as Authelia. A key concern is how authentication would work for services with their own apps, such as Vaultwarden for Bitwarden or Immich. Access control also needs a clear method, such as IP rules, MAC address rules, SSO, or another setup.
The goal is to keep health data from Oura and Apple Watch on a local home server instead of relying only on outside services. The setup would collect and manage that data inside a server at home, such as a Mac mini. There is no clear container or repo already found that fits this need well. Practical recommendations are needed for building a self-hosted health data setup.
A Mac mini owner is considering replacing the original Apple SSD with a different SSD that has more storage. The main questions are whether anyone has already done this, whether it works well after replacement, and whether the swap is safe. No specific Mac mini model, SSD model, step-by-step method, or success result is provided.
PrintGuard 2.0 detects failed 3D prints on the device itself. The new version has been rebuilt around one Python engine that runs in both server mode and browser mode. The React screen layer only shows the interface and talks to the engine through JSON commands and events. For self-hosters, the main change is that Docker is now the only supported way to run it. Images are published for amd64 and arm64, including Raspberry Pi 4 and 5. The included docker-compose.yaml also starts MediaMTX, so one command can bring up both the PrintGuard hub and the video streaming server. Cameras now work as network streams through MediaMTX instead of needing broad device access. It can pull RTSP, RTMP, or HTTP video sources, publish the current device camera through WebSocket, or find streams that are already being sent to the server.
NoMAD-Classic updates the well-tested NoMAD 1.2.2 release to Swift 5 and builds it as a Universal binary for Apple Silicon Macs. The original NoMAD v1 is an Intel-only app that is expected to be deprecated, while NoMAD-2 is described as an abandoned beta that is not usable in practice. NoMAD helped macOS users work with company account systems by supporting things like account conversion, screen unlocking away from the office network, and automatic Kerberos renewal. Claude was used for most of the update work, with another model used for some harder bugs. The practical reason was to keep using a Mac as a main work machine in an organization with weak enterprise macOS support and Windows-only job tools.
The University of California has built a low-carbon computing platform from 2,000 retired Pixel phones. The phones are stripped down to their main boards, loaded with a Linux distribution that removes some consumer phone limits, and grouped into clusters of 25 to 50 devices. Modern phone performance cores can match or beat modern server cores on single-threaded work. Benchmark results suggest that 25 to 50 phones can deliver roughly the work of one modern server. Early tests showed that a 20-phone cluster could handle peak assignment submissions for a class of more than 75 students, with grading delays lower than the default AWS backend. A 2,000-phone setup is expected to support about 100 similar classes at the same time. The system will also be used to study phone-based computing at large scale.
A Mac Mini M4 bought on Amazon at launch price is being used with the Ulanzi QT01 case. The case is designed to make the small Mac look more like the 2019 Mac Pro. It protects the Mac Mini and does not seem to block ventilation. It also makes the power button easier to reach. The front panel can be removed, but using a USB hub avoids taking up the Mac Mini’s front ports directly.
A home server often needs a public IP address or router port settings before people can reach it from the internet. The setup here uses a small cloud server, such as a DigitalOcean droplet, as the public entry point. Server software would run on the cloud server, and client software would run on the home server, so traffic arriving at the cloud server’s public IP could be sent straight to the home server. Cloudflare Tunnel can do part of this job, but it does not freely expose every public-side port in the same way. Pangolin was identified as a tool that fits this need well.
Cisco Unified Call Manager is being installed as a virtual machine on unRAID, but the setup runs into trouble almost immediately. A GitHub guide and a YouTube video helped the system boot, so the first startup step worked. The next screen then shows an error, blocking the installation. The confirmed facts are the target software, the unRAID virtual machine setup, the external install-image guide, a successful boot, and a failure right after that.
PPPC-based management of Mac Accessibility permissions is marked as deprecated. Accessibility permission is often needed by remote-control tools, automation tools, and apps that need to operate the Mac on behalf of the user. The available item does not include the macOS version affected, the final removal date, or the replacement method.
The current setup is simple. An i5-12400 mini PC runs Debian, with Docker Compose handling services such as Immich, Jellyfin, Pi-hole, WireGuard, and Home Assistant. A separate consumer router handles the home network, and the whole setup already works well. The proposed change is to replace the Wi-Fi M.2 card with a 2.5GbE wired network adapter, giving the mini PC two network ports. Proxmox would run on the mini PC, with an OPNsense VM acting as the main router and firewall. A Home Assistant OS VM and an Ubuntu VM would also run on the same machine, with the existing Docker services moved into Ubuntu. A 2.5G switch would be added for the home network side. The main concern is whether this makes a working setup too complex, and whether an M.2 2.5GbE adapter is reliable enough for always-on router use.
The Jellyfin media library uses mostly high-quality video files, so the files are larger than needed for an iPad screen. For a long trip, smaller iPad-friendly copies would make it possible to store more movies and shows for offline viewing while keeping the original library unchanged. The current workflow is manual: pick a few files from the Jellyfin library, then copy or upload them over Wi-Fi into VLC on the child’s iPad. That works for one or two videos when the files are already a reasonable size, but it does not scale well for a larger batch. The desired workflow is to select a folder, playlist, or group of files and automatically transcode them into a smaller iPad-friendly format. Possible settings include 720p or 1080p resolution, H.264 video, AAC audio, and a lower bitrate. Tdarr is being considered, but it may be too complex because the goal is not to reprocess the whole library or maintain a full transcoding pipeline. A simpler option is an ffmpeg script that only converts selected files.
A Mac Mini can slow down when large video files and photos have to be pulled from iCloud during editing work. Moving phone footage through Apple Photos and then into a project folder can quickly consume the Mac Mini’s storage space and system resources. When big clips are imported into iMovie, the app may become sluggish or refuse a new project until older footage is deleted. The proposed fix is to keep media files on a local NAS instead of relying on cloud storage. Active editing files are cached on dual NVMe drives, and the NAS connects over a 2.5GbE network. With that setup, local file transfers can reach up to 300 MB/s, so the Mac Mini does not have to wait for cloud downloads before editing can continue.
The first goal was to run AdGuard Home for home ad blocking and connect it to Unbound for local address lookup. The setup then expanded into password storage, outside access, uptime checks, speed tracking, an ebook library, downloads behind a VPN kill switch, and a Minecraft server. The services include Nginx Proxy Manager, a Let's Encrypt certificate, Vaultwarden, Tailscale, qBittorrent, Gluetun, Uptime Kuma, Netdata, Speedtest Tracker, DiscoPanel, Playit.gg, BookOrbit, and Homepage. Each service gets its own subdomain under a personal domain. The machine is a Lenovo ThinkCentre M720q Tiny with an i5-8500T processor, 24GB of DDR4 memory, and a 256GB M.2 NVMe drive for the operating system and containers. A 512GB SATA SSD is ready to be added for more storage. A small desktop can grow from one simple home-network job into a broad personal server setup very quickly.
The current setup uses a Mac mini to run qBittorrent, Jackett, and Mullvad, with qBittorrent bound to the VPN. Files are downloaded on the Mac mini first, then moved to a Synology DS920+ NAS where Emby uses a specific folder layout. Nothing automatically downloads new movies or TV shows, so searches still happen manually through Jackett. The goal is to simplify the system by handling downloads and organization directly on the NAS with Sonarr, Radarr, and the wider arr stack. The NAS is running DSM 7.3.
The setup uses an M4 Mac mini with 16GB of memory and 512GB of internal storage. The like-new Mac mini cost $750 and included an under-desk mount. AppleCare still had 9 months left. A lightly used 2TB Samsung T7 external drive was added for $250. The Mac mini is connected to an older 27-inch Thunderbolt Display. The 512GB model was chosen because the 256GB model has weaker read/write speeds. The move was from an M1 MacBook Pro with 16GB of memory to a cleaner, more future-ready desktop setup tucked under the desk.
Ironsmith is an open source macOS app that can create personal Mac apps from a prompt. Its goal is to work with small, limited local models such as Apple Foundation models and the Gemma series. The demo video used GPT 5.4 mini because a local model would take too long on video, but the same app was also made with a small Gemma 4 E2B or E4B-class model. The main method is a custom agentic loop designed for small models with limited context. The model generates the whole app in one pass, then many formatting checks, linting steps, and fixed repair steps are run until the result can compile. Small models can write full simple apps, but they need a lot of help fixing made-up details and syntax errors. Very simple apps can be created fully on device, even on a Mac with as little as 8GB of memory. Stronger models still give better apps and fewer errors.
A company is giving away old desktop PCs for free. Each machine has a Ryzen 7 2700 processor and up to 32 GB of DDR4 RAM, but no storage drive. The machines feel too useful to throw away, but there is already a NAS at home running the needed services. Their electricity use could be too high for always-on server use. Selling the parts is also an option if the employer allows it.
A personal self-hosted tool is needed to manage two everyday tasks in one place: work hours and recurring subscription costs. The work-hours side should let someone log hours or tap clock in and clock out from a phone, then check at the end of the month whether the paycheck is correct. The expense side should track regular payments such as Netflix, Amazon Prime, Spotify, internet service, and similar monthly costs. Large business systems such as Odoo, ERPNext, and Jira are too much for this need. Standard to-do lists and Kanban boards are also not a fit. Team management and multi-user permissions are unnecessary because this is for one person. The setup should work well with Docker and be easy to use on mobile, either through a polished PWA or a native app connected by an API. If one app cannot cover both needs, a modern two-app setup with a similar look and feel would also work.
The current setup has a Raspberry Pi 3B, a Raspberry Pi 4B, an external 3.5-inch hard drive, an SSD in a docking station, and a basic TP-Link switch with about 6 or 8 ports. A small, neat rack is needed to hold the gear in one place. The docking station may be too large to fit inside the rack, so placing it on top is acceptable. The practical need is simple physical organization for small server devices, storage, networking, and cables.
kube-coder runs a browser-based development workspace on Kubernetes. It includes VS Code in the browser, persistent terminals, a browser running inside the pod, AI helpers such as Claude Code and OpenCode, and a dashboard. On a Mac, minikube can create a single-node Kubernetes cluster on the local machine and run one kube-coder workspace pod inside it. The setup does not need a cloud account, image registry, DNS, or TLS, and the dashboard is reached at `http://kube-coder.local:8080/` with the default login `admin / admin`. The local commands are wrapped as `make local-*` targets and only talk to the minikube context, which lowers the chance of accidentally changing a real cluster. The workspace image is large, about 2 to 3 GB, because it bundles code-server, Claude Code, OpenCode, Ante, LibreFang, ttyd, and a full toolchain. The first build can take a while, while later runs use cached work. minikube starts with 4 CPUs and 6 GB of memory, so Docker Desktop needs at least that much assigned. Apple Silicon Macs are supported natively on arm64.
A new home server setup now needs a music player that the family can use. The main requirement is playing music through a Bluetooth speaker. It would also be helpful if phones and Windows computers could reach and control it. Several options have already been found, but the goal is to pick a sensible starting point based on real use before installing many different programs just to test them. The practical need is a self-hosted music player that is simple enough for home use.
A Windows 11 Pro home server is being considered for a move to Linux. Windows was chosen because it was familiar and easy, but update and reboot annoyances have become tiring. The preferred replacement is not a command-line-only server, but a Linux system with a normal desktop interface that is easy to learn. Linux Mint and Pop!_OS are being considered. The current machine has an Intel Core i5-12600K, 32GB of memory, built-in Intel graphics, a 1TB NVMe drive for the operating system, and two 12TB hard drives using NTFS. It currently runs Plex, Transmission, and ProtonVPN, with plans to add media automation tools such as Sonarr, Radarr, Lidarr, Readarr, Bazarr, Prowlarr, Overseerr, Jellyseerr, and Notifiarr, plus Jellyfin, Navidrome, and Audiobookshelf.
A free older computer works normally after testing. It has an i3-4330 processor, 12 GB of DDR3 memory, a 180 GB SATA SSD, and several network ports. It was going to be thrown away, but it can still be useful. The likely plan is to turn it into a firewall for a home or small office network. It can be managed through a serial connection, and its front display may be used to show status information with OPNSense. Similar machines sell locally for at least 100 euros, so getting it for free makes it worthwhile.
A very small 7-inch rack was able to hold both a UniFi Cloud Gateway Ultra and an Arris SB8200 cable modem. Others expected the gear would not fit, but the two devices were squeezed into the compact space. The build is not finished yet, and the screws do not all match. JetKVM is being considered as the next possible addition.
A bare metal server from OVH in Singapore is available for one unused month. It has a Xeon E 2136 processor, 6 cores, 12 threads, and 64 GB of memory. Proxmox is already running on it, so it can host several separate virtual servers or services. The server costs 80 USD per month, and the month has already been paid for even though it is no longer needed. Possible uses include running Nextcloud, moving some DigitalOcean Droplets onto it, or running a Tor node for a public network service.
There is interest in linking several Mac Minis together to run automated build work. The goal is a build farm automation agent, meaning a setup that can create and process software builds without manual steps each time. No detailed hardware list, network setup, storage plan, or operating method is included. The main point is using Mac Minis as a small group of work servers, not just as one standalone home server.