“Open source is free” is true the way “a puppy is free” is true. The license costs nothing; the keeping of it does. Here is the honest ledger for self-hosting a business app.
The money
A small production server runs $6 to $48 a month depending on CPU and memory — the same machine comfortably runs one business app plus its chat agent. Compare that to SaaS per-seat pricing: a CRM for ten people routinely costs $300 to $900 a month, forever. On a two-year horizon, self-hosting typically wins by a wide margin, and the gap widens with every user you add.
The time
Someone has to install the app, point a domain at it, renew certificates, apply updates, verify backups, and watch disk space. Skilled, that is an hour or two a month per app. Unskilled, the first install alone can eat a weekend. This cost is why “self-hosted” and “abandoned instance” so often appear in the same sentence.
The attention
The expensive part is the context switch: the upgrade that lands mid-project, the backup you should test but never do, the security advisory for a package you did not know you were running. Two hours a month is realistic for one app; five apps is not five hours, it is a part-time job, because each one has its own stack, release cadence, and failure modes.
What breaks the ledger in SaaS’s favour
If your team is large but your needs are shallow — you just need a chat tool, not a chat platform — per-seat SaaS can be cheaper than your time is worth. The same is true if you have no one who can be on call when the server misbehaves during a deadline.
What keeps it tilted toward self-hosting
Fixed cost that ignores headcount, data you can actually export, and no roadmap roulette. And the operations cost has a third answer now: managed self-hosting. With OpenSysLab the app runs on your own private server — updates, backups, and monitoring handled, starting at $6 a month — which removes the time and attention lines while keeping the money line intact. Add the Vibe-code agent for the chores that remain, and the ledger rarely favours renting.