I just returned to DEVONThink after years (I’ve been using the old versions) with a purchase (upgrade pricing even after skipping v3 which is friendly) of v4.
I already regret that I abandoned it for so long.
For me the AI help with finding and sorting is very useful.
In combination with Bookends where I keep my large database of scientific literature (see e.g. here) and linking to and from Obsidian are very helpful.
4 Likes
Or, I think better, one group with those labels or other meta data used as you want describing each document. That is what i would do.
But then i also do not rely fully on DEVONthink for all my writing and editing. I do store most things in DEVONthink (unless Scrivener stuff which goes to Dropbox and is never indexed into DEVONthink)…
2 Likes
I think you need to try the editor out, really. It is improved to the point that the need for external editors has reduced greatly, but whether you’d want to use it routinely for publishable articles, I’m not so sure, because it doesn’t have the final compilation and publishing features that other programs such as Scrivener, Ulysses and iA Writer do. Not saying it’s impossible, just that this isn’t what the software is aimed at providing.
You could certainly fit it into a workflow with something like Pandoc at the other end very nicely though.
2 Likes
I checked my upgrade price from v3 and it will cost $99 but no indication of what the yearly upgrade will cost. Or will it cost the same $99? If so, that’s a bit steep. Did I understand correctly?
The yearly upgrade is $99 but you don’t have to take the update every year - your existing version will carry on working.
1 Like
For many that would be a problem, but because I use Squarespace, I’m not able to publish directly from Scrivener, Ulysses, or iA Writer; I always have to copy and paste my articles into the SP editor. It’s unlikely that I’ll use DT for writing—it is not designed for that purpose, but I’ll give it a try at least. 
DEVONThink’s built-in editor won’t replace Scrivener or Ulysses. Don’t even try.
The DEVONthink editor is for editing files not managing writing projects like Scrivener and Ulysses.
I don’t use iA Writer but my understanding it is an editor for individual Markdown files, which is the same function of the DEVONthink editor.
However, it’s a personal choice iA Writer and/or/both DEVONthink editor and even then one can store the Markdown files edited with iA Writer inside DEVONthink and use DEVONthink’s “open with …” feature.
4 Likes
If anything, I see DEVONthink’s editing efforts as response to Obsidian.
FWIW I’ve paid for the upgrade and I the price is fair. However, I’m underwhelmed by the new features. I hope that continous development process means they can deliver more interesting features with less wait.
If anything, I see DEVONthink’s editing efforts as response to Obsidian.
I’m not sure where you’re getting that idea but it’s not the case.
I’m underwhelmed by the new features… they can deliver more interesting features
And just what would be “more interesting”?
1 Like
Jim - I posted here and not on the DT forum because I wasn’t trying to be antagonistic. Over the years I’ve used DT3 I’ve been happy enough. I paid for my upgrade because I want you to stay in business. I don’t want to sound like I’m griping.
- Obsidian - your editor may not be a reaction to Obsidian, however if I were to characterize to someone, it has similar degree of polish. In Obsidian the markdown editor is small part of something bigger. With iAWriter et al, the markdown editor is the core thing.
DEVONThink wish list
- Simplify/Overhaul the UI - there is enough complexity in that UI I don’t what to expect. An example navigating from one place to the next, my view changes. I wish it were always consistent.
Search
In my ideal world the search model would semantic. Ex. I just searched “dedicated team membership” - The top result: “Interruptions in Agile Software Development Teams” - DOI: 10. 1177/ 8756 9728 21991365.
It didn’t the things that mattered on this subject:
- “Multiple Team Membership: A Theoretical Model of Its Effects on Productivity and Learning for Individuals, Teams, and Organizations” - MIT Sloan School Working Paper 4752-09
- “Pros and cons of dedicated teams” - Pros and cons of dedicated teams | Dave Nicolette
I realize that on a technical level what I want is very very hard. Unfortunately cloud based tools like NotebookLLM are starting to do this better than DEVONThink. So when I have a needle in the haystack problem, sometimes I try throwing masses of data into NotebookLLM.
I want DEVONThink to succeed.
1 Like
These things are better expressed on our forums. As a production manager of mine years ago always said, “I’m not responsible for things no one tells me about.” Very true in all facets of life, actually. 
And we’ve known each other long enough, we wouldn’t think you’re being antagonistic or even discouraging.
Unfortunately cloud based tools like NotebookLLM are starting to do this better than DEVONThink.
DEVONthink is not trying to compete with bespoke AI applications nor has it become solely “an AI application”. The core focus of the app remains the same and access to external AI is a complement to our own AI, in ways that make sense in our app. We also have data integrity and privacy concerns such apps don’t have to consider, especially given the massive volumes of data people put into their databases. There is a ton of data in many peoples’ databases they do not want exposed to AI in any way (and some who want nothing to do with AI in general).
In my ideal world the search model would semantic. Ex. I just searched “dedicated team membership” - The top result: “Interruptions in Agile Software Development Teams” - DOI: 10. 1177/ 8756 9728 21991365.
This isn’t an effective search since it contains no context. Search what for that phrase? The contents? The name? Tags? Custom metadata?
An example navigating from one place to the next, my view changes. I wish it were always consistent.
It’s unclear what you’re describing here, but if you’re referring to the item list showing as icons, list view, or columns whe you select different locations, that was addressed years and years ago. Set the view you prefer, then enable the General > Interface > Retain view setting. It will remain with the view you chose.
And again, this discussion is better suited in our backyard, especially as development and our other customers can weigh in on the matter.
3 Likes
Yes, we are well aware of embedding and vector databases. However, it’s no trivial matter to just implement such a setup, especially at scale.
I currently have 472 articles on my site. I calculated the 1,536 dimensional embedding vector (array of floating point numbers) for each of those articles, and stored those vectors in my site’s SQLite database.
No disrespect to the author of the article, but look at that number of articles… then consider how many documents are in your databases. I have a very limited number of databases open right now and have ~28,000 documents. And no, it would not be useful for all 28,000 to be embedded any more than it’s useful to dump hard drives worth of documents into a database injudiciously. Just having a ton of data does not lead to it being effective and useful; to the contrary, in many cases. These are things we have to be very thoughtful about.
2 Likes
Please feel free to lift this whole conversation to the DT forum. Three things to consider:
- DevonTHINK isn’t competing with the AI products - ok. Except when an AI product with much better search is released, then they takeout a large chunk of your market by accident. Example: I worked for software company Andyne Computing. We said, we don’t compete with Cognos. It didn’t matter, eventually we were squashed. It was En Passant.
- Metadata - I don’t see how that would help. I need the broadest search to find papers or writing that might relate this search. The amount of effort to have tagged my entire database is huge. So I really on simple searches. When I load 50 documents in NotebookLLM and a similar search, it does a better job of finding related content
- Embeddings Database - this might not be as large as claimed.While main databases are very large, the information density is lower than plain text. A multi-megabyte PDF only has a few hundred kb of text. Audio files → transcripts → another few hundred kb.
Maybe there is a different approach than embeddings. My goal is DevonTHINK handle conceptual search and not just keywords. I’m betting in the next few years, I will get a tool that does this. I hope it’s DevonTHINK.
3 Likes