Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.xpertai.cn/llms.txt

Use this file to discover all available pages before exploring further.

Database resources are suitable for helping Agents understand the structure of SQL DataSources, but they do not encourage Agents to freely write SQL. Through ontology graphs and read-only actions, UOSE turns database exploration into a controlled process.

​
Use Cases

Database discovery is suitable for tasks such as:
  • Find a business table.
  • View table fields, types, and comments.
  • View primary keys, foreign keys, constraints, and indexes.
  • Preview a small amount of data.
  • Explain a SELECT query plan.
  • Execute row-limited read-only queries.
  1. Register an xpert_database resource.
  2. Set catalog/table allowlists.
  3. Enable metadata enrichment.
  4. Complete full sync.
  5. Search tables or columns in the ontology workspace.
  6. Use database.describe_table to inspect fields and constraints.
  7. Use database.preview_rows to preview a small number of rows.
  8. Use database.explain_query to inspect the query.
  9. Use database.query_select to execute read-only queries.

​
Understanding Table Structure

Agents should first read:
  • table name.
  • column name.
  • data type.
  • nullable.
  • primary key.
  • foreign key.
  • index.
  • comment.
  • row count estimate.
This information helps Agents determine join conditions, filter fields, and query cost.

​
Read-only SQL Boundary

database.query_select allows only one read-only statement. Recommendations:
  • Set an explicit limit.
  • Select only necessary fields first.
  • Avoid full scans of large tables.
  • Explain complex queries first.
  • Do not execute DDL, DML, CALL, or multiple statements.
If users request update, delete, or table creation, v1 should explicitly reject the request or explain that the capability is only defined in the ontology and is not executed at runtime.

​
Metadata Enrichment

After metadata enrichment is enabled, the system supplements according to the database dialect:
  • constraints.
  • indexes.
  • foreign keys.
  • check.
  • row count estimate.
If a dialect does not support enrichment or enrichment SQL fails, sync should not fail as a whole. Basic metadata still enters the graph, and the failure reason is recorded in semantic context or abnormal information.

​
Security Recommendations

  • Set strict allowlists for production databases.
  • Limit maxQueryRows.
  • Do not write sensitive connection information into the graph.
  • Display query results minimally.
  • Configure stricter policies for tables that contain sensitive columns.