The honest cost comparison is not “free self-hosting vs. paid cloud.” It is a transparent per-seat licence running on your own compute, set against a cloud bill that charges you per seat and meters every token — with both lines rising as a deployment succeeds. Which one wins depends on how many people use it and how much they ask of it.
The two cost shapes
Metered cloud charges for every seat and then meters every query on top — and as adoption spreads, both the seat count and the token bill climb, with renewal increases layered over them. A self-hosted per-seat licence holds a transparent rate you can predict: the same €25/user/month as your team grows, graduated down at real volume, with inference running on infrastructure you already pay for — or free, locally, via Ollama.
What belongs in a fair comparison
Count everything on both sides.
| Self-hosted total | Metered cloud total | |
|---|---|---|
| Recurring | Transparent per-seat licence (€0 Community / €25 user/mo Enterprise Edition, lower at volume) | Per-seat fees across everyone with access |
| Variable | Infra (modest) + inference on your compute, or €0 local via Ollama | Per-token metering + overages that grow with query volume |
| Increases | Fixed rate within a term, volume discounts as you grow | Annual per-seat increases + usage-tier jumps |
| Operations | Backups, monitoring, updates (your team) | None (fully managed) |
| Risk column | None — data stays in your stack | New sub-processor, transfer risk, un-auditable answers |
The simple test
Estimate where adoption and query volume land in two years, not where the pilot starts. Price the cloud option at that number, add the annual increases and metered overages, and compare it to a transparent per-seat licence plus your operating cost — with local models taking inference to zero. For most enterprise rollouts in DACH, the durable answer favours the stack you control — and that is before the compliance column is even weighed. See the pricing page and Why self-host your RAG platform.
Frequently asked questions
Is self-hosting really cheaper?
It depends on scale and usage. Metered cloud can win at genuinely small, contained, low-query scale with non-regulated data. Self-hosting wins as users and query volume multiply, because your per-seat rate is fixed, inference runs on your own compute (or free locally via Ollama), and the sovereignty and audit benefits come bundled rather than as paid add-ons.
What hidden costs does metered cloud hide?
Per-token overages as query volume grows, annual per-seat increases, and the harder-to-price items: a new sub-processor to assess, data-transfer risk, and answers you cannot fully audit. For a regulated buyer those are part of the real cost.
Sources & further reading
- RAGSuite — pricing (Community / Enterprise Edition) — €0 unlimited users · €25/user/mo, lower per seat at volume
- Self-hosted vs. hosted RAG — the broader trade