Data Catalog¶
Every item of data on the platform is registered in the Catalog: it has a clear location and a record of its source and destination. AI locates data here before retrieval; when you trace an answer’s source, you follow these records back.
Your data |
AI participation |
|---|---|
Every item is registered as workspace → Catalog → database → table / volume, with sources and destinations recorded as it is processed and used. |
AI locates data through the Catalog before retrieval; answer tracing follows the same records. |
Data ownership hierarchy¶
Level |
Description |
|---|---|
Workspace |
Isolation boundary for resources and permissions; members, data, and costs are scoped to a workspace. |
Catalog |
Top-level unit for organizing data assets in a workspace. |
Database |
A grouping in a Catalog that contains tables and volumes. |
Table |
Structured data that physically exists in the MatrixOne data engine. |
Volume and files |
Unit for managing unstructured files. File contents are stored in object storage; a volume registers names, locations, and statuses. |
The distinction between tables and volumes matters: table data is in the database engine, while file contents are in object storage. The Catalog registers both consistently, but their physical locations differ. This determines the processing rule: workflows pass file references, not file content. See Data processing.
Beyond registration: source and destination¶
Data lineage. The relationship between processing outputs and source files is registered when processing finishes. Any text chunk in a knowledge base can be traced back to its source file.
Consumption records. You can trace which knowledge base used data and which Q&A response cited it through lineage. The basis for a knowledge base answer being traceable is here.
Limitation¶
If you modify databases or tables directly in MatrixOne outside the platform, Catalog registration can differ from the actual state. Run metadata synchronization to align them.
Next steps¶
What you want to do |
Go to |
|---|---|
Process registered raw files into retrievable data |
|
Look up exact term definitions |
For AI agents: use Catalog registration to locate data. Tables physically exist in MatrixOne and file contents in object storage. Metadata synchronization fixes registration drift caused by operations outside the platform.