My Obsidian tags were nested. The nest is broken into pieces. Many of the pieces are extraneous. Now I have too many tags. But, I cannot do anything about it because tags in DT are virtual. Anyway, most of my DT vault lacks good tags.
I need more informative titles, metadata might help, the Table of contents of books need to be featured, and Comments are placed in the sidebar. I need to figure out how to word those comments. Maybe AI could help me by looking at texts and suggesting comments. What else can I do to improve my success with AI in finding useful texts? Probably I should put long books (pdfs) in the appropriate folders. Even divide books up so they fit the topic of the folder. Tell AI that this little pdf is an important table of contents for xxx book. My vault is for medical research. How could I make it more AI friendly? Without using up all my free time?
What does that mean – they are no longer nested in Obsidian? What happened to them?
What does that mean?
Which table of contents of which books needs to be featured where?
Perhaps providing more detail, like example tags and what you’re actually doing with Obsidian and DT might help to help you. But as it stands, I don’t understand most of your post.
Sorry. Virtual is the word that AI used. It means that the tag file is not a normal file. You cannot make a folder to sequester excess tags. Ai says that too many tags is not helpful. So I wish I could make my tags fewer and more useful to AI.
The tags were nested and very functional in Obsidian. Now I transferred the notes to DT, and they look very lame to me. I do not see my nested tags. Only fragments.
I have whole books in pdf form. They are Materia Medica. A kind of dictionary of medicines. So there are many entries describing individual medicines. I cannot imagine AI going through those books to locate a medicine that I am researching. But AI could go through a table of contents. Especially, I suppose it would be helpful if AI knew that the TOC was related to the book that I put in the folder.
Ai listed titles of files and tags as being important to find information. Also, comments where I might note what I think is important in the document. I think AI could help me with comments as well as summaries. But if I want a real usable summary, I think I have to export the document and ask an external AI to do the summary. Usable means I can really find out what is in the document. Not just topics.
Metadata seems to not be so powerful, according to AI.
Finally, AI is telling me to put smart folders inside important folders. Folders with subfolders and subsubfolders are not good. Too deep.
I have not been able to use smart folders to find information. First, the information can be buried in a folder, or somewhere in a long paper. The keyword is highlighted yellow, but scrolling down a long paper for one yellow highlight is very time-consuming. The highlight comes up if there is only one. But the others will be buried. I am probably not using DT search correctly. The examples on YouTube are so simple. They are not like my complicated files. I am not keeping track of taxes, or bills. I am looking for complex ideas and topics that can be referred to by many names.
Like @chrillek I’m not sure what help I can be, but to try to get some clarification about what you are doing and where you obtained your advice to help others help … Can please can I confirm:
You “transferred” your Obsidian vaults into DEVONthink. Would be good to explain how you did that. And is the Group structure you have in DEVONthink as created by the import of Obsidian vault or did you change things?
You are using “AI” and YouTube videos to instruct you how setup DEVONthink? What “AI” are you using, and are you sure it (and YouTube) is providing good advice?
Have you reviewed the DEVONthink Manual, especially the chapters “Getting Started” and “In & Out* to get some ideas how DEVONthink works and using the right nomenclature? The DEVONthink Manual is a terrific resource and should be your first stop. After that, consider reading “Take Control of DEVONthink”. Both available for “free”, i.e. paid-for by DEVONthink. See their Support page on the web site.
The guy you need might be Kourosh Dini (Course & Books | Kourosh Dini). He is a medical doctor and a user of DEVONthink. I know he has also used Obsidian as he posted about it on his blog (some time ago).
@rmschne AI online is my source. My experience with forums is that experts are all so far ahead of me. Computer terms make little sense to me. (If you asked me about my specialty, I would be no different. It is hard to talk to the ignorant masses.) I feel accused of not reading the manual well enough. It’s true, so that hurts. I just do not retain what I read. With YouTube, at least the screen is visible. But, the situations are not easy to relate to.
At least, by questioning AI, I can ask the same stupid question many times. I ask until we are communicating. No one is offended. ChatGTP is terrible because I use the free model that does not remember who I am and what I am doing. I have paid to train Claude to understand what I am doing and what I need, but Claude is a poor source for understanding apps. I agree with your probable assessment. I have started checking ChatGTP when Claude gets stuck. Also, I warn Claude not to use script and rely on the functions of the app. But that is beyond Claude’s scope often. Nevertheless, as I say, AI does not feel offended if I ask dumb questions. In the digital jungle, I need a guide to talk to. I can think better when I am allowed to be at my own real level. Probably the best strategy is to talk to AI until it rephrases my question to a more understandable level. Here, I mistakenly expressed myself without asking Claude to speak for me. Big mistake. Sorry.
I have tried these many times. I guess many is not enough. I have to use them more like an encyclopedia. Look, every time I am lost, which is most of the time.
I tried out the DT icons on the left column. I see that they are very informative. There is an AI button. I seem to have signed up for Open AI. I would prefer to use Claude because I am rarely asking questions about the digital jungle. I wonder if I could use the Claude Sonnet 4.5 that I paid to understand my large project?
I want to give AI instructions to summarize materials in a specific folder with a regular set of instructions. For example: lecturer, number of lecture in a set of lecture notes, lecture title. Also give me a brief summary of the lecture with the first sentence describing the whole lecture. If pathologies and medicines are mentioned, I want those to be listed.
I wish folders could also contain such instructions defining the contents of the folder.
No one is offended and we hope you’re not either. However, external AI doesn’t really know anything and says reasonable sounding things that can be inaccurate.
That is one of the reasons why development built the Help Assistant into DEVONthink’s help. It draws on DEVONthink focused information, curated for it, to increase the chances of returning more accurate replies.
Searching within PDF files is a trivial task for DEVONthink. I do it constantly. What do you hope that AI will do that DT’s standard search can’t?
(Incidentally, DT can also break imported PDF books into smaller files, using the ToC as a guide. This is very helpful if you want more precise search results.)
I’m wrong about nothing coming up. If I click on the icons in the left column of any document, I get many connections between notes. I found a poorly titled file. But, if I correct that kind of mistake without fancy adjustments, the relevant files should show up.
I still think I could be more AI friendly going forward. I am summarizing seminar notes using AI. Those files should be made AI friendly.
Titles show the most important information.
The first line summarizes the text.
Maybe I add context for nuances that concern me, but are not necessarily the point of the lecture. I am not sure where I put that.
Tags are imperfect, but as long as I can find my files, I am happy.
Even books are indexed with the context of the keywords. This is what I need!
In my hard disc, I have many files that I do not put into DT. I am not sure what is too much for my main Hardisc, so I am conservative. Is there a way to find out how much space a file is taking? If I were sure that I could quickly find them, I would put all of my essential books into DT. I should not store them in the main HD.
I see. My nested tags should be intact. I should be going to this obvious source of information first. Thank you. I seem to quickly get tired of reading digital information, like eating too much. So, I should take small portions in one meal.
Download a PDF copy of the DEVONthink manual so you’ve got a “hard copy” you can refer to. If you want, you can then add highlights and post-its if you need to. Personally, I keep this outside DT but it’s up to you.
The appendix entry in the DT manual lists all the available Boolean operators, so you only need to learn what they are and how to deploy them correctly.
While I’m not a medical person, I am a scientist, and seems to me like learning how to use DT’s search functions will take you most the way there with your research, you don’t need language models for that (after all, DT has been used for this for many years, long pre-dating the release of language models!).
If you want to use LLMs, it makes sense to bring them in once you’ve got specific files or a group of files to interrogate.
Regarding file size, the info tab in DT’s Inspector will show the size of a group or an individual file.
There is more than one way to look at file size. Space is allocated in chunks on many filesystem types, so a file containing a single character probably might eat 4k of real disk space. I think APFS is an exception to that. If you were to save files to a NAS, the physical space used would depend on how the NAS allocates space.
Tags have always fascinated me, particularly since when I first started using Devonthink I didn’t see the point.
Tags in Obsidian are keywords added to documents. Obsidian can search for tags as text. They are things that physically exist in the files, either in YAML or as text like #tagme.
Tags in Devonthink are groups. The difference between a group we call a group and a group we call a tag is in the meaning, and there are certain rules DT applies to the groups used for tagging.
A Devonthink tag isn’t something added to a file, a tag is a bucket where replicants or subtags can go. The act of tagging a document creates a replicant in the group we think of as a tag.
