Same here. I actually never looked them directly up, if not for backup or maintainance (but always missed my nvim editor).
Personally I don’t take Markdown or text files as requirement for data sovereignty, as long as the note app provides some form of export to file exchange formats or uses a well known database like SQLite, which allows to export notes via SQL.
Out of interest (and not entirely off topic), why are you looking at SilverBullet as a LogSeq replacement? I hear there’s some people unhappy with the whole move to using a DB on the backend, but that doesn’t seem to be the driver in your case?
This is tricky since there’s no folding support in the live preview renderer at all right now…
Some thoughts that came to my mind, roughly in descending order of importance (sorry for wall of text):
Yes, actually I liked this step first, in hope for better performance. It doesn’t feel right to directly write into an outliner-managed file, so DB probably might the better storage for these kinds of notes anyway. But then, there are doubts about vendor lock-in (Logseq became venture capital funded) and uncertainty, when DB version actually will be released. Last, it’s my impression, plain file storage is more beneficial for local/offline AI assistance.
On this topic, @chabbi covered it pretty comprehensively. If I may add my own thoughts on top of that:
Logseq has had such solid ideas (some of which came from other PKMs) and has truly changed note taking for me
Why SB in particular? Again, @chabbi covered it well, but:
One of the technical goals of SilverBullet is that it should be a project that, complexity-wise can still be managed by just me.
I’ve pushed the new linked mentions and new block-aware snippet extraction on main, it’s now on the edge builds. You’ll have to run a Space: Reindex to reindex everything to see the results though.
I agree with all your points, but this one confused me. Can you elaborate? I want to make sure I’m not missing something.
In my view, the Entity-Attribute-Value (EAV) approach to data modeling makes querying much more flexible in Logseq, versus the relational structure in Silverbullet.
In Logseq I can query items based on their relation’s attributes or even their relation’s relation’s attributes (the datalog video I shared covers this well).
As an example, “give me all TODO items that reference the tag #topic, but only those that reference ‘People’ with the page property ‘team’ = ‘Engineering’”
I agree, that’s a good video, and so is your example. What I was thinking when I wrote that was more general than queries (sorry I didn’t word it well), and around easy “user-space” customization:
space-lua is excellent. I’ve been able to implement things around folding, journals and suchFor me, I tend to use the things in the list above far more than the expressivity of querying that Datalog can offer. But even with the relational model SB uses around querying, there is a lot of hackability in post processing relational query results. I imagine your example can be implemented by filtering in post processing, even if the resulting code is not as concise as in Datalog.
There are certainly many entire query classes that can simply not be done at all without EAV, but I’ve not come across too many of these in the PKM context. This might well be indicative of my own constrained thinking. I’d love to learn how you use EAV based querying in the PKM context. Perhaps some illustrative examples would help. Perhaps I’m not using it as well as I could be!
I am a Logseq → SB user, and did so for these reasons:
Works great (side note: ./silverbullet upgrade-edge is awesome!).
One thing: alignment is off. The children are “outdented” while they actually need to be indented wrt to the parent:
Ehm, I’m not sure I understand what I’m looking at. When I try what you’re describing I get the children indented.
and on the sup page linked mentions:
That talk indeed is an excellent introduction of Datalog.
I agree with you, it is a beautiful language for graph queries, especially to find recursive relations of entities (objects, tags etc. in SB). It’s also fun to use, when you got the hang out of it. Imagine tags “SQLite”, “Relational database”, “Database”, and there is need to query backlinks for all kinds of databases, without knowing these relations before-hand. Datalog makes this possible in a concise way. In (some?) SQL dialects there is WITH RECURSIVE CTE like with newer SQLite versions, but I did not find that as elegant - also not sure, if SB supports this?
I don’t know what Logseq actually did to Datalog or its DataScript dialect - but performance for recursive queries was atrocious! I had luck, when it did crash the whole app right away, instead of freezing it for some time 
. Then I gave up on complex queries like these, despite need for it …
Formatting on my end is off in several ways, no idea why:
I’m using Chromium on Windows 11.
Edit: Formatting is perfectly fine on Edge/W11. Strange, SB formatting is otherwise all fine on Chromium.
I did something to make linked mentions kind of work for items like tasks. The slash command copy generates a uid and assigns an attribute and tag to the item. The tag is copied to clipboard and can be used wherever the item is mentioned.
function get_attr(text, attr)
return string.match(text, "%[" .. attr .. ":%s?([^%]]*)%]")
end
local function random_alphanum(len)
local chars = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
local result = {}
for i = 1, len do
local idx = math.random(#chars)
result[i] = chars:sub(idx, idx)
end
return table.concat(result)
end
slashCommand.define {
name = "copy",
run = function()
local line = editor.getCurrentLine()
local temp = editor.getCursor()
editor.moveCursor(line.to)
local uid = get_attr(line.text, "uid")
if uid == nil then
uid = random_alphanum(5)
editor.insertAtCursor("[uid: " .. uid .. "] ", false, true)
end
local tag = "#uid-" .. uid
if string.match(line.text, tag) == nil then
editor.insertAtCursor(tag, false, true)
end
editor.moveCursor(temp)
editor.copyToClipboard(tag)
end
}
Tasks and progress which could be scattered in space.
Tags collect all the linked mentions
I’m guessing some CSS tricks could hide the uid attribute and just show a minimal button instead of the full tag.
It might be possible to make a plug by extending this instead of changing the core functionality of SB unless I’m missing something.
I would also appreciate if someone could point me to a way to sort the "Objects tagged with " section by creation date, last modified or date header, link or ilink (if available) ?
I love the new ‘eye’ icon for transclusions, makes it faster to go to them.
If there is a way to override its behavior, we open the page in the right side panel instead. Making it even easier to quickly change its contents and go back to where we came from. Which makes it closer to ‘live editing’ like in Logseq.
That would be quite a project. It would be more feasible to maybe let e.g. Ctrl or Cmd-clicking opening it in a new tab or browser window.
