Core Concepts
OKF retrieval vs vector RAG
Which questions each kind of retrieval answers best, and when to link both to one engine.
Vector RAG and Docklee OKF retrieval are good at different questions. Vector RAG matches meaning across loose text. OKF retrieval follows how things connect. Most real knowledge benefits from both, and one engine can read both at once.
Where each one wins
| Question | Better with | Why |
|---|---|---|
| "How does A relate to B?" | OKF | It follows typed links. |
| "Who is affected if A changes?" | OKF | It walks several links in one search. |
| "Which tables hold customer data?" | OKF | It filters by concept type. |
| "What's the current policy?" | OKF | It drops superseded concepts. |
| "Find anything like this paragraph." | Vector | It matches by meaning across loose text. |
| "What did the manual say about X?" on long, unstructured text | Vector | Long prose doesn't break into clear concepts. |
When content has no real structure, vector search is the simpler choice, and OKF adds little. OKF pulls ahead as soon as relationships, types, or freshness matter.
Side by side
| OKF | Vector | |
|---|---|---|
| Stores | Concepts and typed links | Chunks and embeddings |
| Dimension | None | One per container |
| Ingest cost | Higher. Building concepts and links takes model calls. | Lower |
| Query cost | Near zero for structural questions | One embedding per question |
| Shows its reasoning | Yes, as a path of links | No |
| Portable | Yes, as an open bundle you can export | No |
Use both
Link vector sources and OKF sources to the same engine, and every question searches both. The results are merged into one ranked list. See Vector and OKF together.
A common split:
| Content | Container |
|---|---|
| Help center articles and manuals | Vector |
| Data catalog, policies, org structure, product catalog | OKF |
Next steps
You know where OKF fits. Turn your content into concepts next.