Memory graph
The structured, queryable map of how your company actually works.
The graph is not a wiki and not a CMDB. It is the operational spine that connects who does what, on which system, with which dependencies, governed by which policy.
Entities
Each entity has a stable identifier, a type, attributes, and a provenance chain back to the artifact that introduced it. Custom entity types can be defined per workspace.
| Type | Note |
|---|---|
| Person | A named individual at your company or an external counterparty. |
| Team | A functional unit. May own systems, workflows, and customers. |
| System | A software service, database, integration, or hardware asset. |
| Workflow | A multi-step process with actors, inputs, outputs, and edge cases. |
| Decision | A recorded choice with context, options, and rationale. |
| Customer | An account, with sub-entities for opportunities and renewals. |
| Vendor | A third-party provider tied to one or more systems. |
| Policy | An internal rule with effective date and approvers. |
| Risk | An identified continuity, compliance, or operational risk. |
Relationships
Relationships are typed, directed, and carry provenance. Edges can be weighted (e.g. strength of a dependency) and time-bounded (e.g. a successor designation that expires at handoff).
| Type | From | To | Note |
|---|---|---|---|
| owns | Person | Team | System | Workflow | Designates accountable owner. |
| depends_on | any | any | Directed dependency; powers continuity risk. |
| escalates_to | Person | Team | Person | Team | Authority chain for decisions. |
| succeeds | Person | Person | Captures intended successor for handoff. |
| informed_by | Decision | Person | Workflow | Provenance for a decision. |
| applies_to | Policy | Team | Workflow | Scopes a policy. |
| exposes | System | Risk | Open risk tied to a system. |
Permission-aware traversal
Every read traverses the graph under the requester's effective role. Edges that cross a permission boundary aren't silently dropped — the response includes a redactions array describing what was filtered, so downstream agents can reason about the gap rather than fabricate around it.
curl -X POST https://api.mnemos.ai/v1/graph/traverse \
-H "Authorization: Bearer $MNEMOS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"start": { "type": "Person", "id": "ent_01HZX..." },
"edge_types": ["owns", "depends_on"],
"depth": 2
}'Ontology evolution
The default ontology is sufficient for most operational use cases. Workspace admins can add custom entity types, custom relationship types, and per-type attribute schemas. Schema changes are versioned; older edges remain queryable under the schema version they were created against.