Most of what an AI needs to be useful at your company is knowledge you already have. The meaning of a metric. The way you write a cold email. The steps to roll back a bad deploy. The problem is rarely that the knowledge is missing. It is that nobody knows where it lives. Half of it sits in a wiki that went stale last spring, some in a Slack thread from March, the rest in one engineer's head.
Google recently published a small spec that takes this problem seriously. It is called the Open Knowledge Format, or OKF, and the interesting thing about it is how little it asks of you.
What OKF actually is
OKF is a folder of Markdown files with a bit of YAML at the top. That is the whole idea. One file describes one thing: a table, a metric, a runbook, a tone-of-voice guide. The only field the spec requires is type. Everything else is up to you.
A file might look like this:
--- type: runbook title: Roll back a deploy tags: [ops, incident] --- # Roll back a deploy If the latest release is bad, revert to the previous tag. See [the deploy pipeline](runbooks/deploy-pipeline.md) for context.
Files link to each other with ordinary Markdown links, so a reader can follow a thread from one document to the next. That turns a flat folder into a graph of related knowledge without any special tooling.
The design choice underneath all of this is that OKF is a format, not a product. There is no SDK to install and no account to create. A bundle of OKF files is readable in any text editor, renders on GitHub, and lives next to your code in version control. A human reads the same file an agent does. No translation step in between.
Where Werkstube comes in
A format on its own does not solve everything. Files in a repo have no access control, no way to keep secrets out, and no answer for "is this still current." They also do not reach the AI your team actually uses. Someone still has to hand the right file to Claude or Cursor at the right moment.
That gap is the part Werkstube fills. You bring your knowledge in an open format, it lives in Werkstube, and it reaches your AI over MCP with the access rules and freshness a team needs.
In practice that means you can point Werkstube at an OKF bundle and import it. Each document becomes a skill. Standard fields like type and title map onto Werkstube's own structure, and anything non-standard in the file is kept exactly as written, so an export later loses nothing. Your AI client then reads those skills through a single MCP connection, and only the people allowed to see a given skill ever do.
The link graph carries over too. When your AI loads a skill, Werkstube also hands it the list of skills that one links to, as references rather than full text. The AI can follow a link and load the next document only if it needs to. You get the structure of a knowledge graph without paying for the whole thing in tokens on every request.
Why the open part matters
The reason to care that the underlying format is open is simple. If your knowledge only exists inside one vendor, you are stuck there. Because Werkstube reads and writes OKF, the door stays open in both directions. You can import what you already have, and you can export the whole workspace back into plain files whenever you want.
So the short version is this. You probably wrote most of these documents already. OKF gives them a shape an AI can read, and Werkstube makes sure the right AI reads the right one at the right time. Nothing about that requires you to start over.