File-System Behavioral Traces for Agent Personalization

4,333 QA pairs are provided across 4 evaluation tracks for grounding memory-centric personalization. The dataset includes 20 user profiles defined by 6 behavioral dimensions and 32 workspace bundles with multimodal files. It was created by Choiszt and updated in April 2026.

Use Cases

• Training personalization models using the 6 behavioral dimensions from the 20 user profiles.
• Evaluating agent performance on tasks defined across the understand, create, organize, synthesize, iterate, and maintain categories.
• Benchmarking memory-augmented models using the 4,333 provided QA pairs.
• Simulating user interaction within the 32 multimodal workspace bundles for context-aware task execution.

Strengths

• Contains 4,333 structured QA pairs for evaluation.
• Defines 20 distinct user profiles across 6 behavioral dimensions.
• Covers 32 task definitions spanning 6 high-level categories.

Limitations

• Dataset size and specific row counts for core components are not provided.
• The scope and diversity of the 32 workspace bundles are not detailed.

Provenance

Source

Hugging Face dataset by author Choiszt.

Collection Method

Constructed as a framework for evaluating memory-centric personalization from file-system traces.

Time Range

null

Freshness

Last updated in April 2026.

Geography

null