Each page is a shortlist of what genuinely replaces a recurring bill — with the licence trap, the migration cost, and a straight note on when the incumbent is still the right call.
per-seat, per-month, billed per collaborator
Most teams pay per seat so that people can send HTTP requests and keep a shared collection. Both open-source options below do that for nothing. You give up the hosted collaboration layer — mocks, monitors, shared workspaces — which is the part you are actually paying for.
metered per task, so the bill grows with usage
Zapier is priced per task, which means the bill grows exactly as your automations get useful. Self-hosting swaps that for a server that costs the same at 100 runs as at 100,000. Below roughly twenty automations the hosted option is still cheaper than your own time.
per-host, plus metered ingest — the ingest is usually the surprise
If what you actually do is look at CPU, memory, disk and containers to find out why a box is unhappy, you are paying platform prices for a dashboard. If you correlate traces with logs across services, you are using the platform and should keep it.
metered per search and per record, so traffic spikes cost money
Self-hosting turns a bill that scales with traffic into one that scales with RAM. That is a good trade until your index stops fitting comfortably in memory, or until you need multi-region latency you do not want to operate.
per-viewer or per-editor seat, which is what makes it expensive once everyone wants a dashboard
Per-viewer pricing is the thing that hurts: the moment dashboards are useful, everybody wants one. Self-hosting removes the per-reader cost and replaces it with a real system to deploy and keep upgraded. Under about ten people, price the hosted option first — your hours cost more.
per-seat, per-month, whether or not a given seat is used
Per-seat pricing is a bad fit for a company where a few people use AI constantly and everyone else uses it occasionally. Self-hosting converts that into a metered API bill plus hosting, which wins above a certain headcount and loses below it. Do the arithmetic on your actual usage before switching.