lin onetwo

weaviate is not ok, it only supports linux https://weaviate.io/developers/weaviate/installation/embedded

Milvus Lite can be used in python https://milvus.io/docs/milvus_lite.md#Configure-Milvus-Lite

Maybe https://github.com/asg017/sqlite-vss is the best choice for embedding vector database

https://github.com/belladoreai/llama-tokenizer-js/issues/5#issuecomment-1592971359 Need to use things like Word2Vec

### Description So searching and auto-generation works better. This can be download on demand, and can be reused when tidgi updated. So the model file can be exist on user/document...

https://github.com/Embedding/Chinese-Word-Vectors

Use llama-node embedding https://llama-node.vercel.app/docs/backends/llama.cpp/embeddings

应该作为一个插件，提供 indexer

语义图片搜索，如果嵌入了 sqlite-vss 向量数据库，可以参考 https://github.com/EdVince/CLIP-ImageSearch-NCNN 对图片都生成向量缓存起来，方便直接搜图片内容

https://github.com/crazko/krystal/blob/master/plugins/rmnvsl/krystal/references.tid