Use these paths when your data is already structured:
| Pattern | Approach |
|---|---|
| Product catalogue, policy fields | Upload JSON/CSV as documents, or POST /facts with infer: "triples" |
| Stable external keys | entity type + name in the unified graph; scope paths for tenancy |
| Graph edges between records | relation records between entities |
Documents vs structured facts
Unstructured text (PDF, Markdown, HTML):
POST /api/v1/{ctx}/documents- async chunking and extraction.Structured triples you already trust:
POST /api/v1/{ctx}/factswithinfer: "triples".Conversational extraction:
POST /api/v1/{ctx}/facts/batchoragent-memory remember.
Send records that already have a structure as triples. Writing them out as documents spends a language model call to recover structure that the records already held. Triples or documents explains the choice, and the sections after it cover validity ranges, registering a scope and batch size.
Cross-layer linking
See Cross-layer linking.