ibl.ai gave every agent a privacy policy setting, shipped in v2.12.0 on 12 October: an admin can let users decide, make every conversation with that agent incognito, or never allow incognito, alongside the existing PII filter. One API endpoint resolves the agent policy, the organization setting, the chat toggle and the user default, and returns the resulting mode, which tier decided it and whether it is locked.
Our readA bank can force an applicant intake agent to keep no chat history or memory while an internal research agent keeps it, instead of one switch for the whole organization. That per agent retention record is what a GLBA safeguards or data minimization review asks to see.