> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getcollate.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Explore the Knowledge Graph in 3D | Collate
> Navigate an asset's neighborhood in the 3D Knowledge Graph, levels, relationship filters, coverage gaps, depth, and derived ontology relationships.
# The 3D Knowledge Graph
Every table in Collate carries a **Knowledge Graph** tab that renders its neighborhood as an interactive 3D force-directed graph. It is the fastest way to see, rather than query, how one asset sits in the wider catalog.
The tab is **hidden unless RDF is enabled**, and is marked **Beta**. It reads from `GET /api/v1/rdf/graph/explore`, which is admin-only.
Drag to orbit, scroll to zoom, click a node to select it. The graph starts focused on the asset you opened it from.
A caption above the canvas states the current level, the node count, and what that level shows, for example *"Data Asset Level · 125 nodes · Tables, columns, owners and queries, and the lineage between assets."*
## Levels
The **Level** control changes what kind of thing the graph is about. It is not a filter. It is a different question.
| Level | Shows |
| ---------------- | ---------------------------------------------------------------------------------- |
| **Data Asset** | Tables, columns, owners, and queries, and the lineage between assets. The default. |
| **Data Product** | Data products and how they depend on one another and share business concepts. |
| **Domain** | Domains and how they relate through shared data products and business concepts. |
Zooming out from asset to domain is the difference between "what feeds this table" and "which parts of the business are coupled, and through what."
## Relationships
The **Relationships** control changes which *kind of edge* is drawn:
| Option | Shows |
| ------------------- | ---------------------------------------------------------------------------------------------------------------- |
| **All** | Both technical and ontology relationships. |
| **Knowledge Graph** | Only structural relations, containment, lineage, ownership, tagging. |
| **Ontology** | Only assets connected through their glossary-term relationships, with the assigned concept laid over each asset. |
The **Ontology** option is the interesting one. It hides the plumbing and draws the business model *as applied to real data*: two tables appear connected not because a pipeline links them, but because the concepts they carry are related. Each node shows its primary glossary term (with a `+N` badge when it carries more), and the caption summarizes `N linked assets · M derived relationships`.
### Derived Relationships
Ontology-lens edges between two assets are **derived**, inferred by overlaying glossary terms, not read from a direct technical link. Selecting one explains exactly how it was derived:
| Kind | Meaning |
| ------------ | ------------------------------------------------ |
| **Same** | Both assets are mapped to the same concept. |
| **Siblings** | Both concepts sit under the same parent concept. |
| **Subtype** | One asset's concept is a subtype of the other's. |
| **Related** | The two concepts are joined by a typed relation. |
The panel shows the full derivation chain as names, for example `savings_account → Savings Account → Accounts → Checking Account → checking_account`, so a derived edge is never a black box.
## Coverage Gaps
The **Highlight coverage gaps** toggle marks, in red, tables with **no glossary-term mapping at all**.
This is the most actionable view in the product for a governance team. It answers "where is my ontology *not* applied?" visually, on real assets, instead of as a number on a dashboard. The gaps are where an AI agent will guess, because nothing has told it what that table means.
## Other Controls
| Control | Effect |
| ------------------------------ | ---------------------------------------------------------------------------------------------------------------------- |
| **Show columns / Show fields** | Expand the focus asset's columns (or data-model / message-schema fields) as nodes. The label adapts to the asset type. |
| **Depth** | Traversal depth, `1`–`5`. Higher fetches more of the neighborhood; the tab opens at depth 1. |
| **Reset view** | Re-fit the camera. |
| **Export** | Download the current scene as a PNG. |
| **Legend** | Node types and link kinds, with colours (bottom-left of the stage). |
| **Fullscreen** | Expands the stage and re-fits the camera (bottom-right of the stage). |
Selecting a node opens a details panel with its type, description, mapped concepts, and neighbors, with **Focus on node** to re-center. Selecting a link opens an edge panel describing the relationship.
## Reading the Picture
Node color encodes entity type and node size encodes importance at the current level; the primary type for the level is emphasized so the thing you asked about stands out. Technical links are drawn blue, ontology links amber, coverage gaps red.
When the server had to bound the result, the caption says so explicitly, *"Showing a partial graph, some relationships were omitted for performance."* A truncated graph is labeled rather than silently misleading.
## Troubleshooting
RDF is disabled on this server, or you are not looking at a table. The tab is registered for tables and hidden when `rdfEnabled` is false.
RDF is enabled but the graph has not been populated. Run the **RDF Knowledge Graph Indexing** application from **Settings → Applications**.
The browser does not support WebGL, or it is disabled. Check `chrome://gpu` or the equivalent, and hardware-acceleration settings.
Drop **Depth** to 1, switch **Relationships** to Knowledge Graph or Ontology to halve the edges, and turn off **Show columns**.
## Exporting the Underlying Graph
The scene export produces a PNG. To take the *data* with you, use the export endpoint the tab reads from:
```bash theme={null}
curl "$COLLATE_HOST/api/v1/rdf/graph/explore/export?entityId={id}&entityType=table&depth=2&format=turtle" \
-H "Authorization: Bearer $TOKEN"
```
`format` accepts `turtle` or `jsonld`. `entityTypes` and `relationshipTypes` accept comma-separated filters.
## Next
When clicking is not enough.
Importance, communities, and paths computed over the whole graph.