FileGram: Grounding Agent Personalization in File-System Behavioral Traces
Shuai Liu, Shulin Tian, Kairui Hu, Yuhao Dong, Zhe Yang, Bo Li, Jingkang Yang, Chen Change Loy, Ziwei Liu
2026-04-07
Summary
This paper introduces FileGram, a new system designed to help AI assistants working on your computer get to know you better and personalize their help, specifically when dealing with files.
What's the problem?
Currently, AI assistants struggle to truly understand how *you* work with your files. They lack enough information to personalize their assistance because collecting data about your file usage is difficult due to privacy concerns and the sheer amount of data involved. Existing AI systems focus too much on what you *tell* them and not enough on what you *do* with your files – like which programs you use, what you edit, and how you organize things.
What's the solution?
The researchers created FileGram, which has three main parts. First, a 'FileGramEngine' creates realistic simulations of how people use files, generating lots of example data. Second, 'FileGramBench' is a way to test how well AI systems can learn from this file usage data, checking if they can accurately reconstruct user habits, understand different workflows, and adapt to changes in behavior. Finally, 'FileGramOS' is a new way to build a user profile directly from the actions taken on files – what was changed, when, and how – instead of just relying on summaries of conversations. It organizes this information into different types of memory, like how-to procedures, the meaning of files, and specific past events.
Why it matters?
This work is important because it provides a way to build AI assistants that are much more helpful and personalized without compromising your privacy. By learning from your actual file usage, these assistants can anticipate your needs and offer more relevant support, making them a more seamless part of your workflow. The researchers are also sharing their tools with the community to encourage further research in this area.
Abstract
Coworking AI agents operating within local file systems are rapidly emerging as a paradigm in human-AI interaction; however, effective personalization remains limited by severe data constraints, as strict privacy barriers and the difficulty of jointly collecting multimodal real-world traces prevent scalable training and evaluation, and existing methods remain interaction-centric while overlooking dense behavioral traces in file-system operations; to address this gap, we propose FileGram, a comprehensive framework that grounds agent memory and personalization in file-system behavioral traces, comprising three core components: (1) FileGramEngine, a scalable persona-driven data engine that simulates realistic workflows and generates fine-grained multimodal action sequences at scale; (2) FileGramBench, a diagnostic benchmark grounded in file-system behavioral traces for evaluating memory systems on profile reconstruction, trace disentanglement, persona drift detection, and multimodal grounding; and (3) FileGramOS, a bottom-up memory architecture that builds user profiles directly from atomic actions and content deltas rather than dialogue summaries, encoding these traces into procedural, semantic, and episodic channels with query-time abstraction; extensive experiments show that FileGramBench remains challenging for state-of-the-art memory systems and that FileGramEngine and FileGramOS are effective, and by open-sourcing the framework, we hope to support future research on personalized memory-centric file-system agents.