About us
We are a studio
with a machine in it
Obsith designs and builds automated systems — the pipeline, the model, and the site that carries them. The work is done by a wired agentic system. Our people supervise it and approve it. Both halves are the product.
What the agentic system actually is
Not a chat window bolted onto a workflow. It is a set of agents wired to real tools — a shell, a repository, a database, a browser, a build — each with a defined job, a defined blast radius, and a written record of what it did. They read the code, change it, run it, and look at the result. Nothing is accepted because it sounds right. It is accepted because it was executed and the output was inspected.
01 Scope, in writing
What is being built, what it is not, and how we will both know it is done. Fixed before anything is made, because a scope agreed afterwards is an invoice, not a plan.
02 The agents build
They work in the real repository against the real tools, in parallel, each with a defined blast radius. Every action is written down as it happens — not summarised afterwards, which is where the errors hide.
03 Everything is executed
Nothing is reported as working because it reads as though it should. It is run, the output is opened, and the number is the number. Where we could not measure something, we write down that we could not.
04 A person approves it
Every consequential change passes someone who can refuse it. That gate is why we can put determinism in a motto and mean it — the same input produces the same result, and somebody signed for it.
How we work
We measure instead of estimating. A claim about performance, cost or behaviour is either backed by a number produced on real hardware, or it is written down as unverified — including in what we tell you. We would rather hand over a smaller scope that is proven than a larger one that is asserted.
What we deliver is yours: the source, the infrastructure definitions, and the runbook that explains them. We do not build dependencies on ourselves.
Bring us the work
Automation of a process that is still manual, a data or machine-learning architecture, or a site built to the standard of the one you are standing in. Tell us what it is now and what it should do instead.