Skip to main content
Platinur’s assistant and model builds run through Platinur’s managed AI (included with your plan) or, if you set one in Configuration → Assistant → Use your own API key, your own provider key — a customer key always takes precedence over the platform default. This page states exactly what leaves your Platinur workspace.

​
What is always sent

  • Table and column names, column types, and null/distinct/row counts from the selected raw schemas
  • Generated dbt model SQL, dbt test definitions, and dbt run/test logs
  • Evidence dashboard source files
  • The instructions and chat messages users write to the assistant

​
Sample values

Profiling also includes up to three example values per column and min/max ranges for time columns, and the assistant may run bounded read-only SELECT queries against the bound warehouse while it works — this is how it learns value formats (date layouts, enum casings, ID shapes) and builds accurate models on the first pass. Columns whose names look sensitive (emails, phone numbers, tokens, and similar) are always excluded from sampling. This behavior is part of the product and is not a per-account setting. If your organization requires a strict schema-and-statistics-only boundary, contact Platinur — a restricted mode exists for enterprise arrangements.

​
What is never sent

  • Warehouse, GitHub, SMTP, or webhook credentials — secrets are stored encrypted in portal metadata and are redacted from logs, status payloads, and support bundles
  • Portal user accounts and password hashes
  • Dashboard render output or exported data files