TechFuel HQWorked examples
These are this calculator’s own outputs, computed with the same formula the tool runs in your browser — so you can see a real answer without touching a single input.
The minimal starter stack a first-time self-hoster builds: Nginx Proxy Manager for SSL, Portainer to manage containers, Uptime Kuma for monitoring, Vaultwarden for passwords. No media, no photos
- Inputs: Checked: Nginx Proxy Manager, Portainer CE, Uptime Kuma, Vaultwarden (4 apps)
- Result: Idle Total 240 MB, Peak Total 430 MB, minimum host RAM (peak + OS overhead) 1.6 GB. All five host sizes pass; smallest fit is 4 GB with 2.4 GB free. Recommendation (green): “This stack fits comfortably on 4 GB of host RAM — 2.4 GB stays free even if every container peaks at once.”
The most common intermediate homelab: reverse proxy plus auto-updates and monitoring, Jellyfin for media, Nextcloud AIO for file sync, Immich without ML for photos, Vaultwarden for passwords. This is the typical “replace Google Photos and Dropbox” build on a mini PC
- Inputs: Checked: Nginx Proxy Manager, Portainer CE, Watchtower, Jellyfin, Nextcloud AIO, Immich (no ML), Vaultwarden, Uptime Kuma (8 apps)
- Result: Idle Total 1.5 GB, Peak Total 3.0 GB, minimum host RAM (peak + OS overhead) 4.2 GB. A 4 GB host is marked insufficient; smallest fit is 8 GB with 3.8 GB free. Recommendation (green): “This stack fits comfortably on 8 GB of host RAM — 3.8 GB stays free even if every container peaks at once.”
How to use this planner
Check the apps you want to run. Idle RAM is what’s used when nothing is happening. Peak RAM is the max during active use — ML indexing, video transcoding, active Office editors. Your host needs to comfortably cover the peak total plus OS overhead.
The per-app figures are sizing bands for 2026 container versions, drawn from vendor documentation and corroborated community reports. I do not run a Docker bench, so they are documented values, not my measurements. Details per app are in their linked tutorials.
One budget line this planner does not include: MCP servers for local-AI assistants, which typically run as bare npx or uvx processes rather than containers. They are not free: the measured MCP server RAM dataset puts each at 178-422MB resident, with a realistic five-server stack totaling just under a gigabyte.
Assumptions and sources
- Per-app idle and peak RAM figures are sizing bands for 2026 container versions drawn from vendor-documented requirements and corroborated community reports. They are not TechFuelHQ bench measurements: I do not run a Docker host, and I would rather label that plainly than borrow authority the numbers don’t have. Your numbers will move with library size, user count, and enabled modules, so treat them as sizing figures rather than guarantees.
- The Immich peak band reflects its machine-learning indexing load; Immich’s own server requirements document the RAM floor it expects.
- The Nextcloud AIO figure follows the RAM guidance in the Nextcloud All-in-One repository, which documents requirements by enabled component (Office is the big one).
- The 1-2 GB host-OS overhead and the 2-4 GB Proxmox hypervisor figure are TechFuelHQ planning allowances, stated so you can adjust them for your distribution and services.