alibaba/zvec — reverse-engineered prompt

Build me a lightweight local vector database that runs inside an app instead of needing a separate server. I want it to feel fast and simple, something I can open from Python and JavaScript, create a collection, define a schema, insert documents with embeddings and text fields, and then search by similarity.

It should support dense and sparse vectors, full text search on string fields, and a hybrid query that can combine vector search, keyword search, and basic field filters in one request. Please make sure the data is persisted safely on disk so it survives crashes, and that multiple processes can read from the same collection while writes stay safe. If there are different index options, include ones that work well in memory and one that is more disk friendly for bigger datasets.

Please also add a small runnable example and basic tests so someone can try it in a minute. If you need details, check the current docs online and make the API feel clean and easy to use.