Datasets & connectors
A dataset is a source for analysis. On creation, the system reads its schema and sample data so the Agent can understand it. Connection secrets are stored encrypted and never shown in plaintext.
Create a dataset
Go to Datasets → New → pick a type → enter connection info → test the connection → save. After saving, the status moves through "syncing → connected"; on failure it shows the error with an explanation.
One dataset can be reused across multiple Spaces.
Supported data sources
Grouped by category (the exact selectable types are shown in the console):
Relational databases
| Type | Key fields |
|---|---|
| MySQL | host / port / user / password / database |
| PostgreSQL | host / port / user / password / database |
| SQL Server | host / port / user / password / database |
| SQLite | database file path |
Warehouses
| Type | Notes |
|---|---|
| Snowflake | account, warehouse, database, schema, etc. |
| BigQuery | project, dataset, credentials |
| DuckDB | database file path |
Lake / table formats
| Type | Notes |
|---|---|
| Iceberg | catalog type, address, warehouse, credentials |
| Delta | table path |
| DuckLake | catalog + data path + object-storage credentials |
| Avro | file path |
Files
| Type | Notes |
|---|---|
| CSV / Excel | upload directly; columns and types detected automatically |
| Google Sheets | sheet URL + authorization |
Object storage
| Type | Key fields |
|---|---|
| S3 (and compatibles like MinIO) | endpoint / region / bucket / keys; switchable vhost / path addressing |
| Azure Blob | account / container / credentials |
Logs / NoSQL / spatial
| Type | Notes |
|---|---|
| Aliyun SLS | endpoint, project, logstore, keys |
| MongoDB | connection info, database, collection |
| Spatial (GeoParquet) | file path |
Improve answer accuracy
- Fill in column descriptions: adding business meaning to fields makes the Agent's SQL more accurate.
- Maintain business notes: write table relations and metric definitions in the Space.
- Re-sync: after connection or schema changes, re-sync so the Agent reads the latest structure.
Troubleshooting
- Connection test fails: check credentials, network reachability, and that the database / table names are correct.
- Abnormal status: open the dataset details for the specific error — often expired credentials or a network issue.