Real-World Deployment Challenges in File-System Event Logging and Privacy Controls

Develop robust OS-level data collection and preprocessing mechanisms for FileGramOS-style deployments that (i) disambiguate and correctly order interleaved multi-application file-system event streams, (ii) deduplicate and reorder duplicate or out-of-order events arising from cloud-synced file systems, and (iii) implement per-directory privacy opt-in and opt-out controls.

Background

The paper outlines a practical deployment pathway for FileGramOS using native OS monitoring APIs (e.g., FSEvents, inotify/fanotify, ReadDirectoryChangesW) and a three-part system consisting of an event collector, an Engram encoder, and an on-demand memory consolidator and retriever.

While mapping OS notifications to the 12 atomic action types is straightforward in principle, the authors explicitly highlight unresolved real-world issues that must be addressed for reliable operation: interleaving across multiple applications, artifacts introduced by cloud synchronization (duplicates and reordering), and strong, fine-grained privacy controls at the directory level.

References

Key open challenges include interleaved multi-application event streams, duplicate or out-of-order events from cloud-synced file systems, and per-directory privacy opt-in/opt-out controls.

FileGram: Grounding Agent Personalization in File-System Behavioral Traces  (2604.04901 - Liu et al., 6 Apr 2026) in Appendix: Discussion and Resources, Deployment and System Integration