Turin, Italy, July 14, 2026. Yempik, a software house specialised in software, automation and AI agents, today published the Company Brain Maturity Model, an open framework that describes how a company’s operating knowledge lives and works with AI across five levels, from L0 (knowledge trapped in people’s heads and scattered chats) to L4 (agents running real work on a knowledge base the company owns). Alongside the framework, Yempik released cowork-os as open source under the MIT license.
The term “company brain” has moved from idea to category in 2026, with investors and AI labs publishing their own adoption frameworks. Yempik’s contribution is practical and opinionated: a way for a team to place itself on the scale, see the cost of staying there, and take one concrete next step.
Most companies do not have a knowledge problem, they have a knowledge-ownership problem. Their operating knowledge lives in people’s heads and disposable chats, so every AI session starts from zero. A company brain is the fix: the decisions, context and state your work runs on, in files the team owns and can correct. We open-sourced cowork-os because the layer that makes this real should be something you own, not something you rent.
What the framework measures
The Company Brain Maturity Model uses five levels: L0, knowledge in heads and chats; L1, written but dead (a wiki nobody updates and the AI ignores); L2, owned and structured (decisions, state and rules in files the team governs); L3, self-updating and queryable (the brain absorbs from where work happens and answers with a source); L4, operated by agents. A seven-question self-diagnosis returns a level and a single next step, not a checklist.
What cowork-os is
cowork-os is an open-source workspace and process layer for Claude Cowork and other file-reading AI agents. Decisions, open questions, project state, next actions and recurring routines live in plain Markdown that the agent reads before it acts and updates when it finishes. It is deliberately not a memory engine: the operating context is on disk and inspectable, outside actions are draft-only, and the team stays in control. It installs into a Claude Cowork project as a one-command plugin or by guided setup, in about thirty minutes, with no training required.
The hard part of agentic work is not memory, it is control. If the context an agent relies on is a black box, you cannot trust it on long, multi-step tasks. cowork-os keeps that context in plain files you can open, compare and correct, so the agent stays grounded and every outside action is a draft you approve. That is a design choice, not a limitation, and it is why the approach holds up in real production work.
Yempik runs its own company on cowork-os, from marketing and website work to sales follow-ups and recurring automations, and ships the framework, the workspace and a complete sanitized example openly so any team can adopt the method.
Availability
cowork-os is available now on GitHub and on its project page. The Company Brain Maturity Model is published on yempik.com and linked below.
The Company Brain Maturity Model
The five-level framework (L0 to L4) at the center of this release, with a self-check that returns your level and one next step.
About Yempik
Yempik is a software house specialised in software, automation and AI agents, serving companies in the Italian market, from a working prototype (MVP) to production-ready systems. Yempik’s promise is fixed pricing agreed up front, source code owned by the client, and technical people as your point of contact. Yempik was founded by Raffaele Zarrelli, with Simone Bova as CTO. Legal entity: Yempik Ltd (UK), company no. 16239387. More at yempik.com.
Media contact
Raffaele Zarrelli, Founder, Yempik
founders@yempik.ai · yempik.com