pixeltable/pixeltable ? reverse-engineered prompt

Build me a Python library and local app for managing multimodal AI data in one place.

I want users to be able to create tables for text, images, video, audio, documents, JSON, and embeddings, then insert data from files, URLs, or cloud storage. It should support computed columns that run model calls or custom Python functions automatically when new rows are added, plus views that can split documents into chunks, video into frames, and audio into segments.

Please also include vector search, simple filtering, import and export helpers, a command line interface, and a basic dashboard so people can browse tables and inspect rows. Make it feel like a single backend for building RAG style and multimodal AI apps, with caching, retries, transactions, and good documentation examples.

Use the current project structure, look up any current docs online if you need to, and make sure it runs cleanly with tests.