- Yempik runs its own company on its company brain of files: about 137 markdown files with decisions, project state, rules, and open questions.
- Those files don’t live inside a tool: they sit on our computer and in version control, readable by us and by Claude, updated by habit.
- On the L0-L4 maturity ladder we’re at L4: agents work on the brain and keep it fresh. The company operates from it, not from the head of whoever’s around that day.
- The structure is the same one we give away for free with cowork-os: open source, copyable, zero lock-in. The method we sell is the one we use.
We don’t describe it in theory: we use it every day
It’s easy to write about a company brain without having one. We did the opposite: we built it for Yempik first, then pulled the method out of it. Today our company runs on about 137 markdown files that hold everything needed to work: what we decided and why, where each project stands, which rules we follow, what’s still open. When we start a session with Claude, we don’t start from zero: the context is already there, written down.
They’re not files scattered across a drive. They’re one place, structured, on our computer and in version control, where every decision has a date and a reason, and every project has a state we update by habit. The difference from "we keep it in mind" is that the knowledge no longer depends on who’s in the office that day: it’s readable by anyone, and by Claude.
Before selling it to others, we use the company brain on files ourselves. Yempik runs on 137 files: it’s not a metaphor, it’s where we live the work.
Decisions, state, rules, open questions
We didn’t invent the structure for the occasion: it’s the same one we release for free with the open-source cowork-os kit[1]. Folders for context, decisions, marketing, the website, reviews, with an already filled-in example to start from. Yempik’s 137 files are this structure filled with truth: our decision log, our project states, the rules we work by, the glossary of our terms.
Plus the instructions, the routine automations, and a real, already filled-in example to copy from.
The number grows every week, because the brain isn’t an archive you fill once: it’s context that updates at the end of every task. When we close a task, we note what changed in the right file, so the following week Claude and the team know a little more. It isn’t heroic discipline: it’s a routine, and today’s 137 files will be more next month.
On the L0-L4 scale we’re at the top, and we know what that means
The company brain maturity ladder runs from L0, where everything lives in heads and chats, to L4, where agents run the work on the brain and keep it fresh. Yempik is at L4, and we say it without triumph: it means agents read the files, work on them, and update them, and that the company operates from that context. It’s not an achievement that makes us special, it’s the logical consequence of putting the method before the product.
We care about being honest about what this means and what it doesn’t: L4 isn’t magic, it’s a routine kept alive every day, and underneath it is the same boring work of writing decisions and states we ask clients to do. If you want to know where you are on the same scale, take the 2-minute self-check: it gives a name to your level and points to the next step.
Our brain is on files we own, and the structure belongs to everyone
There’s a reason we can tell all this with no reticence: there’s no secret to protect. The structure of our company brain is public, it’s the open-source cowork-os repository[1], and anyone can copy it and fill it with their own company. Our 137 content files stay ours and private, but the framework belongs to everyone. No vendor in the middle, no proprietary format, no lock-in: if we switch tools tomorrow, the brain comes with us because it’s just text we own.
It’s the same thing we promise the people who work with us, and the proof it isn’t marketing: the method we sell is the one we run our own company on. How you build it, step by step, with interviews, transcription, and validation, we wrote here: how to build a company brain on files. And if you’d rather we set it up for you, with your process and the code staying yours, see our AI agents for companies, with code you own.
The Yempik case, in practice
Does Yempik really use a company brain on files?
Yes. We run our own company on about 137 markdown files that hold decisions, project state, operating rules, and open questions, on our computer and in version control, not inside a vendor. The method we sell to clients is the one we use for ourselves: we built it for Yempik first, then pulled the product out of it.
What does being at L4 on the maturity ladder mean?
It means agents read the company brain files, work on them, and keep them updated, and that the company operates from that context instead of from the head of whoever’s around that day. It’s not magic: underneath it is a routine kept alive every day, the same work of writing decisions and states we ask clients to do. It’s the top level of the L0-L4 scale.
Why exactly 137 files, and what do they contain?
It’s the real count of our workspace at the time of publication, and it grows every week because the brain updates at the end of every task. The files contain the decision log, project states, operating rules, the glossary, and open questions, in the structure of the open-source cowork-os kit filled with our own company. We cite no other numbers: this is the only figure, and it’s real.
Is there lock-in in your company brain?
No. The brain is just text on files we own, and the structure is public in the open-source cowork-os repository, copyable by anyone. There’s no vendor in the middle and no proprietary format: if we switch tools, the context comes with us. It’s exactly the promise we make to the people who work with us, and the reason we can tell all of it without reticence.
Sources
- [1]cowork-os, Yempik’s open-source repository. github.com
This page is written by Raffaele Zarrelli and Simone Bova, founders of Yempik, with editing done with Claude. The company brain and its maturity model are Yempik editorial models. The number of about 137 files is the real count of our workspace at the time of publication and grows over time; we cite no other numbers. The kit cited is our open-source cowork-os (MIT license).
Want to run your company on a brain like ours?
We start from your process, not the tool. We help you put decisions, state, and rules on files you own, governed, with the same open-source structure we use for ourselves. Fixed price and timeline, the code is yours.