Since 2010

Eighty-four projects on our own servers, on open source

Git, continuous integration, image registry, SSO, DNS, metrics, logs, monitoring: all of it runs on our own machines, on open-source components. Not out of principle — because it holds load better, costs a fraction, and does not stop when a vendor changes its mind.

Notre flotte
84

projects versioned, deployed and monitored by the same pipeline

150 / day

production releases, across 65 projects — no meeting, no deployment window

125

services under permanent watch, checked from the outside every minute

238

automated checks re-reading the code of the whole fleet, continuously

The context

An agency selling hosting, continuous deployment and monitoring ought to be able to show what it runs itself. On those three subjects it is the only proof that counts: they cannot be demonstrated with a screenshot from a client, only by what you operate every day. We run 84 projects — 27 of them for paying clients — on our own dedicated servers in France, on open-source components, with the same pipeline for all of them. A new client project inherits, on day one, everything built for the other 83.

What we did

  • check_circleEverything is self-hosted on dedicated servers in France: the Git repository, continuous integration, the image registry, SSO, DNS, databases, metrics, logs, monitoring, the inbound mail server. No hyperscaler anywhere in the chain.
  • check_circleDeliberately simple orchestration that two people understand in full — not Kubernetes. Choosing a tool you have not mastered moves the risk rather than reducing it, and an outage at three in the morning is not solved by reading documentation.
  • check_circleOne delivery pipeline for all 84 projects: one command, the tests run, the image is built, the new version proves it responds before the old one stops, and rolling back takes a minute.
  • check_circle238 automated checks re-read every project's code continuously: vulnerable dependencies, exposed secrets, conventions, dangerous queries, pages missing structured data. What they find becomes a ticket, not an email someone archives.
  • check_circle125 services checked from the outside every minute, logs centralised and searchable over a month, metrics per container and per server. Including the overnight jobs, whose failure raises no visible error.
  • check_circleThe few exceptions are deliberate and written down: outbound mail delivery goes through a specialist provider, because an IP sending reputation cannot be manufactured in-house. Self-hosting is not a dogma, it is a trade-off — and it is justified line by line.

What changed

  • trending_upInfrastructure cost bears no comparison with a public cloud at equal load: on dedicated hardware the same power costs a fraction, and the bill does not depend on this month's traffic. That is what lets us offer hosting to a sole trader and to a group alike.
  • trending_upNo lock-in. No proprietary service you cannot leave, no format you cannot export, no API metering bill that quietly explodes. The day a client wants to leave with their data, they leave with their data.
  • trending_upIt scales by adding machines, not by upgrading a subscription tier. And when a third-party service changes its terms or shuts down, our pipeline does not stop — that has happened to enough others for it to be a criterion.
  • trending_upEvery client project inherits, on day one, everything built for the other 83: backups, HTTPS, deployment, monitoring, alerting. The foundation is never billed twice.

Let's talk about your project

Thirty minutes is enough to tell whether we are the right fit. We reply within 48 hours, and we say no when it is not for us.