Core Concepts
What is OKF
Google defined a format for writing knowledge down. Docklee turns that format into retrieval your app can ask questions of.
In June 2026, Google Cloud published the Open Knowledge Format, an open, vendor-neutral way to write down what an organization knows: its tables, metrics, APIs, policies, and runbooks, as linked Markdown files. OKF standardizes how knowledge is written. It doesn't define how to find anything in it.
Docklee closes that gap. The same way vector RAG turned a pile of documents into something a model can search, Docklee turns OKF into retrieval. You send content, Docklee builds the OKF for you, and your app asks questions through the same calls it uses for any engine.
From documents to answers, two ways
Vector RAG, the approach most knowledge products use today, works like this:
- 01SplitCut documents into chunks.
- 02EmbedTurn every chunk into a vector.
- 03MatchFind the chunks closest to the question.
- 04AnswerHand those chunks to the model.
Docklee OKF retrieval works like this:
- 01BuildTurn content into typed concepts and typed links.
- 02EnterMatch the question to the right concepts.
- 03WalkFollow links up to three steps out.
- 04AnswerRank by authority and return tight context.
Your code doesn't change between the two. You link an OKF source to an engine and call lynk.query or lynk.retrieve, exactly as you would for a vector source. See Query OKF.
Format, retrieval, or both
| Vector RAG | OKF on its own | Docklee OKF retrieval | |
|---|---|---|---|
| What it is | A way to search documents | A file format | A retrieval layer built on the format |
| Turns your content into it | Yes | No. You write the files yourself. | Yes, from documents, apps, and Dock writes |
| Answers questions | Yes | No | Yes |
| Follows relationships | No | Links exist, but nothing follows them | Yes, across several steps |
| Knows what's current | No | No | Yes. Superseded concepts are dropped. |
| Shows why a result was returned | No | No | Yes, as the path of links |
| Needs an embedding for every chunk | Yes | No | No. Structural questions need none. |
| Portable | No | Yes | Yes. Export the bundle at any time. |
The format
An OKF bundle is a folder of Markdown files:
| Part | What it is |
|---|---|
| Concept file | One file per concept, such as a policy, a product, a team, or a metric. |
| Frontmatter | Structured fields at the top of each file: its type, title, description, and tags. |
| Body | The concept itself, in plain Markdown. |
| Links | References from one concept to another. Together they form a graph. |
| Folders | Group related concepts by domain, such as engineering or finance. |
A folder of photos tells you what each photo shows. A family tree also tells you how everyone is related. OKF is the family tree.
What Docklee adds
The format deliberately leaves the hard parts open. Docklee fills each one:
| The format leaves open | Docklee adds |
|---|---|
| Creating concepts from content you already have | Generation from your documents, apps, and Dock writes. See Generate OKF. |
| What kinds of concepts and links exist | A shared vocabulary of 43 concept types and 31 relationship types. See Semantic vocabulary. |
| How far to trust a concept | Confidence, authority, and status on every concept. See Concept files. |
| What to do when concepts disagree | Duplicate merging, conflict markers, and supersession. See Merging and conflicts. |
| Searching a large bundle | Retrieval that finds entry concepts, follows links, and ranks by authority. See How OKF retrieval works. |
| Using OKF for memory, not only documents | Concept graphs built from memory spaces. See OKF in memory. |
Every file Docklee writes follows the published OKF format, so other OKF tools can read it. Docklee's additions sit in their own block that other tools ignore.
Your knowledge stays yours
- Encrypted at rest. Concept files are encrypted with your workspace's key. The searchable index keeps concept bodies encrypted too, and only the few concepts that answer a question are decrypted.
- Scoped to linked sources. An engine reads only the OKF sources linked to it, the same rule as vector sources.
- Portable. Export your bundle as plain files and use it anywhere. See Export OKF.
Next steps
You know what OKF is and what Docklee builds on it. See how retrieval walks the graph next.