Graph Geeks
If you've worked with graph databases, knowledge graphs, or ontologies before, here's how Atlas maps to concepts you already know.
It's a labeled property graph
Atlas is a pre-typed labeled property graph with a defined ontology. Every Point is a node with a type label. Every Path is a directed edge with a type and optional properties. Both nodes and edges can carry arbitrary key-value properties.
| You say in Atlas | Graph people say |
|---|---|
| Point | Node / Vertex |
| Path | Edge / Relationship |
| View | Subgraph / Projection |
| Atlas | Graph |
| Type | Label / Class |
| Properties | Properties / Attributes |
Pre-typed, not schema-free
This is the key difference between Atlas and a raw graph database. You can't just create any node with any label. Atlas enforces an ontology: a defined set of valid Point types and Path types, with constraints on which types can connect to which.
Person → has_role → Position is valid. System → has_role → Metric is not.
This is deliberate. Schema-free graphs get messy fast. A typed, constrained graph is queryable, validatable, and AI-readable. The ontology is what makes this a knowledge graph, not a whiteboard.
Ontology categories
Point types are organized into semantic categories:
| Category | Types | Represents |
|---|---|---|
| Actor | Person, Position, Group, Organization, Vendor | Who |
| Action | Process, Step, Review, Decision, Approval, Handoff, Task | What happens |
| Resource | System, Artifact, API, Equipment, Skill, Transport | What's used |
| Outcome | Outcome, Metric | What's produced |
| Location | Physical Site | Where |
| Time | Date, TimeRange, AbsoluteDate, RelativeOffset | When |
Path types are similarly constrained. reports_to expects an Actor source and target. has_step expects an Action source and Action target. The full ontology is available in the Type Glossary.
Graph algorithms
Atlas supports standard graph analytics:
- PageRank: find the most influential Points
- Betweenness centrality: find bridge Points between communities
- Community detection: clustering via Leiden algorithm
- Connected components: find disconnected subgraphs
- Node similarity: Jaccard similarity between Points
- Shortest path: BFS between any two Points
- Degree centrality: most connected Points (in/out/both)
Available via the MCP server. See the API Reference.
Cypher queries
You can run arbitrary Cypher queries against your Atlas via the query_cypher MCP tool. See the API Reference for details.