Data processing system for identity-based, locally executable control of AI agents with configurable access policy management.

Abstract

A data processing system for identity-based control of AI agents, comprising a processor unit and a non-volatile storage medium, characterized in that the system includes: (a) a local identity file on the user's storage medium, containing cryptographically linked identity bindings with an assigned trust level, a persona configuration structure, and an optional cross-session context buffer; (b) a persona management unit with a configurable set of user role types, comprising at least: • a restricted role type with an activatable content filter level for users with increased protection requirements, • a standard role type, • an administrative role type with extended configuration rights compared to other roles;(c) a locally encrypted data vault unit that stores confidential data end-to-end encrypted on the user's storage medium and makes it accessible to the agent exclusively via opaque pointers, without transmitting plaintext data across network boundaries; (d) a policy enforcement unit with an allowlist configuration structure and an editorial unit that, prior to each agent output, automatically checks whether the output is permissible based on the active policy configuration and automatically suppresses impermissible content, the check unit being terminable in real time for each interaction; (e) a locally executable agent runtime unit with a fully locally executable execution architecture that does not require a network connection to an external service as a prerequisite for core functionality.