Restrict Assistant Responses to Indexed Knowledge Only?

[kreativekrise]

Hello,
is there a way to configure Onyx (formerly Danswer) so that the assistant strictly answers only based on indexed / internal knowledge, without using general LLM knowledge?

Ideally, we would like to:

• Disable any fallback to general LLM reasoning.
• Ensure the model retrieves and answers exclusively from the indexed documents.
• Optionally enforce this behavior per assistant or workspace.

Is such a configuration currently possible (via UI, environment variable, or prompt control)?
If not, is this feature planned or already under consideration?

Thank you for your help!

[antoniosb]

@kreativekrise I guess you can achieve that by setting the DISABLE_GENERATIVE_AI environment variable to true.