What this is.
Compai is the open-source documentation of a real AI operating system: a company Brain that captures approved operating sources, turns them into source-linked memory, and supports verified work through agents, tools, skills and the team's AI clients.
The system, in one paragraph.
A central Brain — structured, versioned and source-linked — sits between the company and its tools. Approved chat, meeting notes, mailboxes, documents and source systems flow through explicit privacy and coverage rules. The Brain formalizes that context into decisions, tasks, a current-state model and health-checked knowledge. Seven production agent runtimes plus a founder command center use 97 authenticated MCP tools and 373 available skills, of which 47 are company-governed canonical procedures. Consequential work writes back a receipt. Context in. Verified execution out.
Who's behind it.
A small team operating one omnichannel consumer business, growing 100%+ a year without an internal engineering department. The system was built by operators with AI tools, in public, with the failures documented alongside the wins. We publish the playbook because the hard part of becoming AI-native is not access to a model; it is the operating discipline required to make context, authority and verification work together.
Why it's free.
Because we wish this had existed when we started. The playbook, the starter kit, the patterns and the lessons are free to use with attribution (CC BY-NC-SA 4.0). Most readers will self-serve. A few want hands-on help wiring it into their company — that's the only thing we charge for, and you can just write us.
Anonymization note: the reference company isn't named on this site. The numbers are real and audited; the brand identity isn't the point.
Where to go next.
The 66-chapter playbook is the core asset. The evidence dashboard shows the dated production snapshot and known gaps. The story is the honest history. The repo is where you fork it.