DTPro/Office: Search unusable

let me be frank: search is unusable in devonthink. it feels very 1999 and gets in the way.

the (temporary?) solution is simple: give me NLP, like almost all apps do nowadays.

case in point: i was looking for an image having the word ‘pew’ in it. this is the workflow:

  • cmd+sift+f
  • enter pew
  • stumble with the mouse to ‘advanced’
  • look for kind (granted, could have been any other name really but i knew it from a previous search)
  • look for photo or image or the like in the list (didn’t know it hence was looking around in that unsorted mess of a list)
  • click ok

now it automatically filters. removing that filter (which isn’t shown anywhere), is equally awesome.

  • cmd+sift+f
  • enter “kind:image pew”

Unfortunately i doubt this will ever make it into DT, you guys seem rather set in your ways.


?? It looks like you just used it. :wink: Couldn’t resist.

I wouldn’t go as far as “unusable” – though I agree the Search panel could be more “modern” (whatever that means) and keyboard-friendly. I usually don’t bother setting parameters because it’s faster to get a pile of results and look through them, than it is to setup the parameters that would narrow my results. I’m lazy. 8)

I don’t see a lot of specific suggestions for improving Search come along in this forum. It’s a complex feature that hints at all of the metadating that’s going on in the internals of DEVONthink as it sorts through your data. This probably is one reason it’s hard to do a UI without all the buttons and settings.

@bosie pointed out one approach or case. I wonder if that will encourage more dialog here.

Like – what would you consider “best practice” in search (NLP was suggested – anything else)?

What apps do search well?

What (other) cases do you run into where DEVONthink Search’s results could improved – or cases that DEVONthink isn’t handling?

I’ve always thought that DEVONthink does an excellent job and finding what you may not know you know (the AI in See Also), but does a less than stellar job of making it easy to locate what I do know that I have. The Global Search does work, but as mentioned it takes so long to use it for complex searches that, as mentioned, it can be frustrating. I’ve also been very vocal over the years about the lack of good search tools for tags, which is the only reason that I index my databases. With indexed documents, and can perform quick, yet detailed, tag searching with a number of third-party apps (Punakea is my tool of choice).

heh, true. but that was honestly forced and every time i use the search in dtpo i hate myself.

thanks for your other reply, have to think about it, a few things i can suggest but no idea about what is ‘best practice’ in that industry. i just know neither evernote (had to mention it to win some friends back) nor DTPO (now i lost them again) have figured it out :wink:

Thinking about this some more – it’s worth asking “who needs all this complexity”? Is Search over-designed? There are at least 60 parameters (counting settings for each option that has settings). Too much? That’s a lot of possible combinations. (OK, this is overstating the results – but you get the picture.)

Maybe what we need is a simple, small interface with limited parameters for most searches. (Think: the menubar “Find” box optionally enabled across all open databases.)

Then, for superusers do an Alfred-like workflow editor with scriptable plugins to build sophisticated search workflows for those who need them.

Just saying…

In such cases, when I know the kind, I typically know the suffix also, so I just search for “pew jpg” (or “xyz pdf”), which works just fine for me.
Otherwise I click on the column for “kind” to get those shown.

Nonetheless …

… this is a great idea. I suggested exactly this in March via email. (Found this quickly by searching for “kind eml” :slight_smile: )

What I would love would be a quick-list of databases in the search window and that the search results would get filtered if I’d choose one. The reason behind this is that I often search for a term, which I want to find e.g. in my business database, but not in any other, especially not in my business-email-database.
I’d still would want to keep the “search in” “databases” option as is now as I can limit the search to folders there, but that list could adapt if one database is choosen in the quick-list mentioned above, so that only folders of that database are shown.

The same list I could well imagine for “kind” as long as “kind: image xyz” is not existing.

All for faster access than with the ways the mouse has to go now.

I find the search not too complex, but in some cases too time consuming. When I’m in a hurry (I have a shop situation) I mostly tend to not limit the search with the given options, but scrolling and clicking through the results and columns always hoping but sometimes failing.

i want the functionality but not the complexity. it is very nice to have or otherwise you might have to go through quite a huge list for no real reason.

yes and no. i am no expert on this but aren’t the parameters also dependent on what you actually search for?
if i search for kind:pdf, can i even search Organization (what is this anyways?)?

Couple suggestions, none is outside the box thinking unfortunately:

  • NLP. i type “tag:foo” and it filters for items having the tag foo. “tag:(foo|bar)” etc
    – might do a shortcut way like “#foo” instead too, to really speed it up :wink:
  • live preview: if i type something, show me the results immediately
  • do away with any popup window
  • cluster the results (e.g. grouped by database)
  • flexible parameters: dont make me think. kind:image is the same as kind:photo as is kind:illustration (not sure about this one)
  • auto completion: when i type “tag:”, give me a list of all the tags
    – for example i want all pdfs with a tag “foo” containing the word ‘bar’. i might enter “bar tag:foo kind:pdf”. but what if i type “kind:pdf tag:foo bar”. after typing “kind:pdf tag:foo” i get completion, the scope of the completion gets narrowed because of the first two things, i might not even be able to enter bar because there is no bar for pdf’s with the tag foo…
  • search on top of a previous search: comes in handy with the auto-completion, since the scope of potential completions is much smaller now (e.g. tags that aren’t in the previous search would not show up in the completion)
  • if using wildcards/operators, show me what i am actually matching with that wildcard/operators.
  • current “search in”: either go through NLP or make it searchable as well. going through the tree takes too long.
  • remove the search icon
  • make me aware that not all combinations of parameters actually work. don’t allow nonsense queries without ‘altering’ the user
  • show me metadata of a search: groups/tags/words/timespans etc. in a sidebar, similar to what thunderbird does. let me filter on that too by clicking on it. download.cnet.com/i/bto/20091208 … 10x578.png (IIRC thats called faceted navigation)
  • want to blow me away? give me this aqua.org/explore/animals#location=.l01 [requires javascript] when i type “kind:image” :wink:

bottom line: i bet one could really do something with the search window, at the end of the day, its half the equation of using a database.

i like that. and alfred’s plugins have taken off (sort of). it hasn’t really happened for devonagent though? think DT will have more success with scriptable search-plugins?


Great ideas. I really like the Aquarium example.

I don’t follow this. The Search panel shows preview results now, doesn’t it?

not if you are in the ‘advanced panel’ (edit: that was not the best phrasing, i was thinking ahead to an NLP-like search where filtering for a tag in the searchbox reveals the result as opposed to now where the advanced panel does not provide such instant feedback)

Hi Devonthink has been amazing for my law practice. Its an amazing bit of software but I agree with these criticisms of the search function. It needs a pretty radical overhaul IMHO.

Why with Boolean searching do not only targets that meet the search parameters become highlighted, but every instance of a word that is part of the Boolean search. So if I search dog NEAR/10 box why does every instance of “dog” irrespective of its proximity to “box” get highlighted - in addition to instances of “dog” NEAR/10 “box”? Thats not the search I asked for so why assume I want to see every seperate instance of dog and box?

And why can I only conduct Boolean searches from the search box and then not conduct same Booelan search using command F once the document has been opened? I obviously want to Booelan search the actual PDF so that I can highlight/circle/comment on the hit.

On an unrelated note why can’t I link across pdfs? Eg I found “dog” NEAR “box” in one pdf deposition - so why can’t I hyperlink it to another pdf where something relevant occurs? At the moment everything has to go through a text note that I can paste a COPY PAGE LINK to.

Again, I think Devonthink is amazing - but the search function needs a big rethink to ensure I can quickly find what I need in my ever expanding information collection.