Custom open-source builds

The rest of this site is deliberately narrow: four categories of thing we have shipped repeatedly and can quote to a fixed price. This page is the honest admission that clients do not always need one of those.

What this is for

Most “we need something custom” conversations turn out to be assembly problems rather than invention problems. The capability exists somewhere in open source — it is just spread across three projects, none of which quite fits, and none of which anybody has joined up for your particular constraints.

That is the work: an internal tool that has to talk to a twenty-year-old system nobody wants to touch; a reporting layer over data that lives in four places; a regulated workflow that no product models correctly; a device or a protocol integration with no vendor behind it; a migration between two things that were never meant to meet.

How it runs

It starts with a discovery week rather than a quote, because quoting bespoke work from a description is how both sides end up unhappy. You get a written recommendation at the end of it — what exists already, what would have to be built, roughly what that costs, and whether it is worth doing at all.

Often the recommendation is smaller than the request. Occasionally it is “buy the commercial product, it is cheaper than us” and we say so in writing. The discovery is chargeable and self-contained, so that advice costs you a week rather than a project.

If we do build, the same rules apply as everywhere else here: open source throughout, hosted in Romania, Sweden or Germany, committed to your repository as we go, and handed over with runbooks. Where we fix or extend an upstream project along the way, we contribute that back — you get the benefit of not maintaining a private fork forever.

What we will not take on

We are seven people with a specific competence, and the fastest way to waste your money is to hire us for something outside it. We will say no to native mobile applications, anything requiring a design team we do not have, machine learning research as opposed to applying existing models, and hardware engineering.

We will also say no when the honest answer is that an existing product already does this well. That happens often enough to be worth putting on the page.