FAQ

FAQ: AI assistance

Looking for the full list of AI features? See Where AI helps in Kenseme.

What if an AI suggestion is wrong?

Reject it, clear it from the review list, or type your own value. Your edits always win. If the Ontology Assistant made a change you don’t want, undo it from the drawer’s Activity tab. See Reviewing AI suggestions.

Does the AI change things without asking?

No AI runs until you start it. Most features propose and wait for you to accept. A few do what you asked straight away, such as generating descriptions or asking the Ontology Assistant to add a class, and those tell you before they overwrite anything and can be undone where noted.

What does the AI see?

It depends on the feature:

  • Schema, Pipeline, and ontology features see the definitions you’re working with: names, types, relationships, descriptions, and any instructions you type.
  • The Ontology Generator also reads the sources you pick, including process documents, glossaries, standards, and the table and column structure of databases and lakehouses.
  • The Glossary Extractor and process documents read the document text you upload.
  • The Data Agent and Ontology Data MCP run read-only queries against your bound database and see a preview of the results so they can describe them.

Can the AI change my database?

No. Nothing in Kenseme’s AI writes to your databases. The Data Agent and Ontology Data MCP only run read-only queries. Everything else works on the definitions stored in Kenseme, and generated SQL is yours to run.

Can I turn AI off?

No AI feature runs unless you start it, so you can leave them unused. Optional features such as the Data Agent, Ontology Data MCP, and the Star Schema designer can also be switched off for the whole organization. See Feature flags.

🚩 TODO: Confirm whether there’s a way to disable every AI feature for an organization at once (there’s no such switch on the Features tab today).

Does AI count against my usage?

Yes. Every AI feature draws on your organization’s monthly AI usage allowance. At 100%, new AI work stops until the allowance is topped up with a usage pack or the next period starts; everything else keeps working. See AI usage allowance.

Are AI descriptions reliable?

They’re a good first draft, not a replacement for review. Read them before you rely on them, especially for business-critical objects.

Why did the AI get my column wrong?

Usually because the name is cryptic and there’s no description. Add descriptions and set naming conventions; every AI feature does better with them.

Can I choose which AI model or provider is used?

Not from the app. The models are set up for your Kenseme deployment by Macula.