Submitted to: Rob LoCascio, CEO, Uare.ai Submitted by: Wolfgang Höltgen, user since the Eternos migration (3 years) Via: Featurebase Feedback Portal The Problem: The current platform offers only two access levels: private (creator only) or public (everyone). There is no intermediate layer. Role-Based Access Control (RBAC) at platform level would require maintaining multiple separate twin instances updated independently, which is not scalable. What Already Exists: My Retrieval Skill already defines seven roles with corresponding filter markers embedded in the training corpus: #Filter_Kind_unter_18_Start/End, #Filter_Familie_Start/End, #Filter_Ehefrau_Start/End, and others. The logic is ready. The platform enforcement is missing. What I Need: A middleware layer between the retrieval process and the LLM that: Reads the authenticated user's assigned role Retrieves chunks from the knowledge base as usual Filters out any chunk containing a #Filter_[Role] marker that excludes that role, before passing content to the LLM Passes only the permitted content to the LLM for response generation_ This is classical attribute-based access control applied to the retrieval pipeline. The filtering logic is a simple string match on start/end markers. One corpus, one twin, multiple filtered views depending on the authenticated role. Business Relevance: This requirement is universal across professional and enterprise deployments. A consultant cannot expose client-confidential data to all subscribers. A CEO cannot grant board-level access to all employees. A doctor cannot give an assistant access to patient records. Without content-level ABAC, the platform cannot serve professional and enterprise use cases. This is a foundational requirement, not a feature enhancement. Requested outcome: Is this architecture on the roadmap? If yes, what is the estimated timeline?

Please authenticate to join the conversation.

Upvoters
Status

Completed

Board

Feature Requests

Date

20 days ago

Author

hoe

Subscribe to post

Get notified by email when there are changes.