DT3 and Zettelkasten

DT3 is the most powerful tool available for Zettelkasten, which is basically a method to write atomic notes and create explicit connections between them.

DT3 has all tools for Zettelkasten: nice text editors (in both markdown and rtf), several types of links that can be used to connect the notes, and backlinks (list of links between notes). In addition, the search is fast, robust and flexible.

What I would like to ask is if someone have ever tried to use DT3’s AI to reveal new connections between notes that had not been identified when writing the note. if so, what was the main outcome.

Furthermore, I want to ask how useful should be for knowledge workers to have in DT3 a tool that builds networks (such as the one available in Devonagent) based on the similarity among notes.

Finally, if this tool become available, should people using Zettelkasten as their approach to make notes stop to making connections between them writing and simply use AI instead to reveal such connections?

I use DT for Zettelkasten as you suggest. But I index files from Obsidian. I will write about my experience soon on my blog, wippp.com

Cheers, Jeff

1 Like

Jeff, excellent! Keep us informed!

I asked a somewhat similar question on MPU last year and got some solid answers. Some were links to this forum and are worth exploring.

@cornchip. Thank you. Very useful!

On a related topic, I’ve always wondered if there are any law enforcement use cases for Devonthink. I don’t think police are as adept as they should be in relating case notes.

In a proper world, skilled gunfighters could justify scoffing at Matt Dillon, but the archivist? Always fear the nerd!

There are less than we’d like as Window$ still runs the show and the inertia in bureaucracy keeps the status quo, but we do have some people in law enforcement using DEVONthink for sure.

On the flip side, I think DEVONthink would be a great tool for organizing white-collar crime. :stuck_out_tongue:




I am using DT3 in one Database as a Zettelkasten. I do activate “See Also” inspector in that mode indeed. It definitely helps as software is able to catch up some suggestions about which I would sometimes not even think about.

With the recent filter on “current database only” you have an ideal setup, which allows to only search for connection within a database - hence a separate DB for Zettelkasten seems to be highly advisable imho.

1 Like

Thanks @jooz. Very useful,

1 Like

I continually use DT3’s AI to guide connection-building between knowledge fragments scattered across multiple databases. Hopefully, some of my observations can be useful to you.

  1. Make liberal use of new databases. When in doubt, create a new database, then close it until you want to use it again. See Also searches open databases: use that knowledge to optimize database size versus scope. For example, I have a database dedicated to researching emerging technology identified by Gartner. When I add and synthesize material related to interesting technologies, I typically keep my professional database open looking for relevant connections. (I am a mechanical design engineer by profession.) Right now, See Also is showing me that an article on autonomic computing is related to some notes from a one-on-one meeting with my manager, and some notes on program management. That’s interesting: what could autonomous computing network models have to do with transmission/power-train design and project management? I’ll just replicate those files into a new folder, replicate other potentially-related documents identified by See Also into that folder, then start to think about new ways to approach my craft on Monday morning. (To DEVONthink devs: thank you for including the Current Database Only option in See Also. Could you also create an Other Databases Only search option as well to increase novelty for workflows like mine?)

  2. Don’t be afraid to exclude documents and folders from Classification and See Also. When I use academic papers for research, I typically keep them in a folder that is excluded from Classification and See Also at least until I’m done reading, summarizing, and exporting the summary as a Markdown document. Bad OCR and extensive citations tend to make entire academic papers more trouble than they’re worth when using the AI, but your mileage may vary based on your needs.

  3. Understand that you’re responsible for interpreting AI suggestions. Sometimes the results will make perfect sense and you didn’t need AI to know the relationship between documents. Sometimes the results will make no sense and you can just go on with your life. But when you’re not sure why the AI thought two documents are connected, you owe it to yourself to spend a little time thinking about why that might be.

If you have any questions, comments, or concerns, please feel free to ask. I really want to become an “AI power user,” so these dialogues are very helpful.


Thanks. Very useful! I hope others will share how they use DT AÍ to find connections among documents.

Interesting; thank you for sharing your approach. My biggest concern with many databases is that managing which are open will become too much of a chore. To what degree have you found that to be the case? Do workspaces or other DT tools help?

Do you ever find yourself merging these granular databases?

Please excuse the response delay: I’ve mulled over your questions and reconsidered some aspects of how I interact with my information. I would also like to emphasize that this is all based on my experience and opinions. Take what you need and please question what you don’t so that I can see where I can improve. (Seriously!)

The plain answers to your questions are that managing a lot of databases isn’t difficult. Databases I use frequently remain open or in the Recent Databases list. They’re typically always open, or two clicks away from it. I do merge databases frequently and delete the empty database. (Right now there are only two databases that aren’t in my Recent Databases list that I feel that I don’t want to delete, but I don’t feel that File > Open Database... isn’t too much effort to manage those.)

On the topic of Zettelkasten and linked thought, I would like to take a moment to thank @ryanjamurphy for his DEVONlink Obsidian plugin. I’ve appreciated your contributions from afar and wanted to take a moment to thank you publicly. :pray:

I have two “wish list” items for DEVONthink devs related specifically to this workflow:

  • Manage database state in workspaces. I’ve tried to set up workspaces with specific databases open (and more importantly, some closed), but they don’t seem to save database state. I tested this by opening specific databases, changing some view options, then saving the workspace. I closed a couple of the previously-opened databases, opened one that I didn’t want in the workspace, and changed some view options. When I invoked the workspace, the view options changed as expected but the databases didn’t change state. If I’m doing this incorrectly or need to change some preferences, please let me know: this would be an extremely valuable feature for me. (And I ain’t too proud to beg!)
  • Increase the scope of See Related Text on selected text. I love this feature, but even if I select All Databases as the search scope, it only seems to return results from the current database. (If this is a bias in my perception, please let me know. It feels like this is a recent issue over the past couple of days, but some important databases have also changed significantly over the past couple of days, so this may be my own doing. I’m currently using version 3.8.2.)
  • Persistent Concordance elimination preset dictionaries, or the ability to eliminate probable junk from Concordance results. It feels like I’ve always spent an inordinate amount of time using Exclude and screening for bad OCR (thanks @Silverstone for the amazing OCR grading script!), yet gibberish persists.

There is something else I’ve found myself wishing for while using DEVONthink: better control over the AI’s scope. (Or a reference to the part of the documentation that corrects my ignorance. :grinning:) The AI is very good at surfacing related content in See Also & Classify, but I would appreciate more control over the scope of the search. For example, I would appreciate an Other databases only check box, the ability to manually specify which databases to See Also, the ability to set a ratio for how many results from any specific source, etc.

Whoops, life calls! There’s always more to discuss, so please poke and prod and I’ll be very happy to reconsider and respond!