Guide
The recipes that run comfortably on the cheapest VPS or an old thin client — what they cost in memory, and where the number stops being the whole story.
A memory minimum is a threshold, not a forecast. It says the recipe was tested with that much and behaved; it does not say the application will stay inside it under any load you invent. The two things that break the estimate are the same everywhere: an import or an initial scan that touches everything at once, and a dataset that grows for years without anyone looking at it.
The ceiling also hides where the real cost lives. Every service here is small in memory and none of them is small in disk over time — a query log, a feed archive and a photo of every scanned page all grow quietly. Read the disk estimate on each page next to the RAM figure; on a 1 GB machine the disk is usually what runs out first.
One practical note about a small VPS: memory pressure on a machine this size does not degrade gracefully. Without swap the kernel kills the largest process, which on a shared box is rarely the one that caused the spike. If you plan to run several of these together, give the machine a swap file and set a memory limit on the container you can most afford to lose.
Sets
The sum of the minimums the recipes below declare. It is a floor, not a recommendation: leave room for the operating system, backups and growth.
Why it is here: 256 MB, and most of that is the runtime rather than the data. A vault for a household is a few megabytes of encrypted blobs; the memory figure does not grow with the number of passwords.
Why it is here: 256 MB for filtering every query on the network. The query log is the part that grows, and it grows on disk — set the retention deliberately.
Why it is here: 256 MB with an important caveat: the initial scan of a large library is the hungriest thing this service ever does, and a first indexing run on a very big collection can want more than steady-state playback.
Why it is here: 256 MB, and the number is driven by how many files are being watched rather than by how large they are. Hundreds of thousands of small files cost more than a few enormous ones.
Why it is here: 512 MB is comfortable for dozens of checks. What eventually grows is the history database, not the resident memory.
Why it is here: 512 MB for the reader and its PostgreSQL together — the lightest way to keep a feed archive that outlives the services it points at.
Why it is here: 512 MB, with the ledger stored in SQLite. The budget of a household is small data by any measure, which is why this fits where a hosted product would have needed an account.