Tasks, recipes and money: which of these you will still be using in six months

Sort them by the habit, not the feature list#

Nothing here fails technically. grocy is one PHP app over one SQLite file, idles under 50 MB, and has an ops load of 1. It is also the most abandoned tool in this category, and the reason has nothing to do with software.

Stock tracking is double-entry bookkeeping for food. Every purchase and every consumption has to be recorded, and grocy cannot know you ate something. The inbound half is a barcode scan; the outbound half is a chore. Miss a week and the pantry claims four tins of tomatoes when you have none, which is worse than no pantry at all, because now the shopping list lies to you too.

That is the real axis for the whole category. Rank by the daily commitment each tool extracts:

• Mealie: paste a URL, get a filled-in recipe. Near-zero effort, immediate payoff.
• Vikunja: add tasks, which you already do somewhere.
• Actual Budget: reconcile weekly against statements.
• Firefly III: learn double entry, then reconcile weekly.
• grocy: record everything you consume, forever.

The other honest constraint is mobile. Self-hosted productivity is weak here and it is the deciding factor more often than any feature. Actual is the strongest because it is genuinely local-first: the budget lives on your device and the server only relays changes, so it works offline. Mealie is a PWA that needs the server reachable. Vikunja reaches phones over CalDAV at /dav, which its own docs still label early alpha, with DAVx5, OpenTasks and Tasks working, and Thunderbird and iOS CalDAV Sync listed as not working. Recurrence rules flow out to clients and never back in.

The default#

Mealie plus Actual Budget, because they have the best effort-to-payoff ratio and both survive contact with a household. Mealie is one container at about 300 MB, imports recipes from a URL and turns a week's plan into a shopping list. Actual is one container at about 120 MB with no database service, on port 5006.

Add Vikunja if you need shared task lists: one Go binary serving UI, /api/v1 and /dav on port 3456, under 100 MB with SQLite.

Two costs to accept up front. Mealie's install docs tell you to set an explicit memory: 1000M limit, because Python pre-allocates larger arenas on big hosts and the container then idles far above what it needs. And Actual's end-to-end encryption password cannot be reset, recovered or brute-forced; encryption is also one-way, and bank sync tokens for GoCardless or SimpleFIN sit on the server outside it.

Two cases where it is the wrong tool. You want a ledger, not a spending chart: Firefly III, with real double entry, multi-currency accounts and import rules, at four containers and a mandatory cron container. You want Trello's drag-and-drop feel for a team: PLANKA, after reading its licence and the 2.2.0 change.

The mistakes#

• Restoring Actual without a sync reset. The server holds change history, so an older restored file syncs straight back to where you were. Restore, then reset sync from that device, then revert the others.
• Letting an Actual budget file grow. Every mutation is stored, so the file grows even when transactions do not, until it hits the 20 MB ACTUAL_UPLOAD_FILE_SYNC_SIZE_LIMIT_MB default and sync starts failing without a size warning.
• Pointing Firefly III at PostgreSQL without changing DB_CONNECTION. It defaults to mysql, and startup fails on a connection error that names nothing useful. Behind a proxy you also need TRUSTED_PROXIES=** or you get http:// links inside an HTTPS site and login loops.
• Upgrading Firefly III past 6.6.0 without replanning integrations. That release invalidated every OAuth token and client, so the data importer and any scripts need new tokens.
• Upgrading grocy and not loading the root URL. Schema migrations run on a request to /, not at startup, so new code runs against an old schema and you get scattered SQL errors. 4.7.0 separately invalidated every session and reorganized AUTH_CLASS.
• Leaving PLANKA's SECRET_KEY at notsecretkey. It is the published example value and it signs every access token. The shipped compose also sets POSTGRES_HOST_AUTH_METHOD=trust. If you are anywhere in 2.x below 2.2.1, upgrade first: 2.2.1 fixed a path traversal that let any authenticated user read arbitrary files.
• Missing Vikunja's VIKUNJA_SERVICE_PUBLICURL. CORS has been on by default since 1.0, so a mismatch gives you a UI that loads and then fails every request.

Read next#

Actual Budget vs Firefly III settles the money question, Vikunja vs PLANKA the task one. Replace Mint or YNAB and Replace Todoist are the migration paths with the export formats named. Then Backups that actually restore, because a SQLite file copied while the container runs is how most of these get lost.