There is a moment in every agency's growth where the software bill stops looking like a rounding error. Ten tools at thirty dollars a month is three hundred and sixty a month, and none of them individually felt like a decision.
The obvious response is to self-host. There is a free, open alternative to nearly everything you are paying for. What nobody puts in the comparison table is what you are actually swapping.
What free actually costs
A subscription is not really buying you software. It is buying you someone else being responsible when it breaks at two in the morning. When you self-host, that person becomes you.
- Updates. Security patches are not optional, and an unpatched service on a public IP is found by scanners in hours, not weeks.
- Backups. Not just taking them. Testing that they restore, because a backup you have never restored is a hypothesis.
- Certificates. Renewal is automatable and still the single most common cause of a self-hosted service going dark.
- Storage and monitoring. Disks fill. Logs grow. Something has to tell you before a user does.
None of this is difficult. All of it is ongoing. That is the part the comparison table leaves out, because it is a running cost measured in attention rather than money.
The four cases where it wins
Self-hosting is not a philosophy, it is a decision with conditions. It pays off in four situations, and they are worth naming precisely because the rest of the time it does not.
- 01Confidentiality. Client contracts, legal documents, medical records and anything under an NDA. If uploading a file to a third-party website would breach an agreement, the hosted tool is not an option at any price.
- 02Per-seat pricing you have outgrown. Tools priced per user punish exactly the growth you are working for. A team of fifteen on a per-seat plan is where self-hosting maths usually flips.
- 03A feature locked behind enterprise. Plenty of tools put single sign-on, audit logs or an API behind a plan priced for companies far larger than yours. Sometimes the self-hosted version simply includes it.
- 04Volume. Automating something a thousand times a month against a metered API costs real money. Running it yourself changes the unit economics rather than the monthly bill.
Start with one, not ten
The failure pattern is predictable. Someone reads a list of self-hostable software, gets excited, and stands up six services in a weekend. Three months later two have stopped working, nobody is sure which, and the backups were never tested.
Pick the single tool where one of the four conditions above clearly applies. Run it for a month. Break it deliberately and restore from your backup before you trust it with anything real. Only then consider the second.
Read the licence before you sell it
This is where agencies get caught. Self-hostable does not mean you may resell it, and the licences differ far more than the marketing does.
- Permissive licences such as MIT and Apache let you use, adapt and build commercial work on top with very few conditions.
- Copyleft licences such as AGPL require you to publish your modifications if you offer the software over a network. That is a real obligation, not a formality.
- Source-available and fair-code licences are not open source at all. They usually permit self-hosting for your own use while explicitly forbidding you from reselling the software as a hosted service.
Using a fair-code tool to deliver work for a client is normally fine. Charging clients for access to an instance of it, as though the tool were your product, is the case those licences exist to prevent. Read the licence file, not the landing page.
The honest summary
Self-hosting is a good answer to confidentiality, seat count, locked features and volume. It is a bad answer to a thirty-dollar bill that mildly annoys you.
The agencies that do this well run a small number of self-hosted services they genuinely depend on, and pay for everything else without guilt. The ones that struggle are running twelve, maintaining none, and calling it a cost saving.
