Access to sys_generative_ai_log for Now Assist analytics and debugging

[Simon Midtvedt]

Hi all,

I’m looking for input from others working with Now Assist on whether you have enabled read access for admins to the sys_generative_ai_log table in your instances.

According to the official documentation for Now Assist Analytics Dashboard Indicators (link here), many of the analytics insights are retrieved from this table.

We are exploring whether accessing this table directly could help us:

• Provide more granular insight into the use of assists/actions at the ticket level (e.g. number of assists per ticket)

• Debug issues with prompt execution, translations, or skill outputs, particularly when working with custom skills or evolving prompts

My understanding is that this table currently has maint-only access by default, and that viewing the logs directly isn’t enabled for admin roles out of the box. Have any of you opened up read access to this table for troubleshooting or deeper analytics? If so, did you encounter any challenges or risks we should be aware of?

Would direct access to this table be a viable path for tracking which assists/actions were triggered for specific tickets, or is that information better retrieved elsewhere?

Thanks in advance for any insights or experiences you can share!

[Wetterhorn]

@Lener Pacania1 Is there any update on when the sys_generative_ai_log table will be made accessible to admin users?

[mattzirp]

Hey folks, not sure anybody bothers to update threads like this but I got here from searching and the information above is not accurate. You can swap out maint for admin on every ACL, but there is an access handler that will still block read of the records on sys_generative_ai_log. 

I spent a while troubleshooting and opened a support case due to the misleading information above, so hopefully someone sees this and saves themselves the trouble. You CAN use sys_gen_ai_log_metadata, but that only presents high level information about the skill/prompt and its success/error. I was looking for more granular info about context of runs that may have gone awry

Support has told me potentially December for this access to be opened, but given the Q3 safe harbor above this may not be a priority to the team.