transformerjs can load onnxmodel from customed url , you can download the model file to local then nextTime without the download step it'll work

transformers.env.backends.onnx.wasm.wasmPaths = '/plugins/SiyuanAssistantCollection/static/';
transformers.env.allowRemoteModels = true;
transformers.env.localModelPath = '/public/onnxModels/';

the model file may be found in this url

async function getModelInfo(modelName) {
    let url = `https://huggingface.co/${modelName}`;
    let response = await fetch(url);
    if (!response.ok) {
        throw new Error(`获取模型信息失败: ${response.statusText}`);
    }
    return await response.json();
}
let modelInfo = await getModelInfo
/**
* then find those files need like model.quantiliazed.onnx | tonkenizer.json and so on 
*/
`https://huggingface.co/${模型信息.modelId}/resolve/main/${文件名}`;

or you can try the method transformerjs used
transformers.js/src/utils/hub.js at main · xenova/transformers.js (github.com)

and to write file into siyuan's workspace，you can try use siyuan's kernelApi

kernelApi.js is just a wrap of siyuan's backend api

SiyuanAssistantCollection/source/polyfills/fs.js at master · leolee9086/SiyuanAssistantCollection (github.com)

SiyuanAssistantCollection/source/polyfills/kernelApi.js at master · leolee9086/SiyuanAssistantCollection (github.com)

use this css snippets to hide tooltips

#tooltip{

display:none 
}

you can open your siyuan in browser with a remote pc ,then you can download plugin and other extension contents

Embedded Block can only use SQL sentence starts with "select * from blocks" for some reason
And dataBase block can not use SQL for now

win server 2019 works fine

select * from blocks as A where A.root="<the page id>" and A.alias

there is a plugin for easier search，maybe you can try that

just use your vps IP, if you dont have a public IP address, try tailscale or zerotier,

You can try using object storage like amazon's S3 or minio;

Previously, SiYuan also tried methods like ，git and other version management solutions., but later changed to the current object storage technology solution because it was not suitable for SiYuan's usage scenarios.

en ，a shotcut of siyuan kernel

in windows maybe you can build a shortcut on your desktop to lunch siyuan without UI

I guess the AttributeView feature under development (testing may start around the end of this month) should partially meet such requirements.

Looking forward to it

However, as far as I know, the performance of Siyuan's listBlock should at least not be much worse than specialized outline software (if not better); the main uncertainty lies in how the performance would be at its limit and whether the functionalities fully meet the needs of habitual users of outline note software.

Actually, I don't use the list function very often. I usually only use it when creating outlines. As for performance testing, I have only done it for normal document formats, and most devices should not experience lag when reading Chinese novels with over one million characters (except for possibly taking a moment to import). I will try to find familiar users who can speak about the performance of lists.