VectorVFS: Your Filesystem, Now a Vector Database

VectorVFS is a lightweight Python library that transforms your Linux filesystem into a vector database using native VFS extended attributes. It stores vector embeddings directly alongside each file, eliminating the need for separate indices or external databases. This turns your existing directory structure into an efficient, semantically searchable embedding store. VectorVFS supports Meta's Perception Encoders (PE), including image/video encoders for vision-language understanding, outperforming InternVL3, Qwen2.5VL, and SigLIP2 on zero-shot image tasks. While supporting both CPU and GPU, embedding a large image collection without a GPU may take time initially. Currently, only Perception Encoders (PE) and images are supported, with more models and data types planned for future releases.