Recall.ai, mymind, read-later apps and junk drawers

[Crystalliere]

Is anyone using Recall.ai?

In late November Michael Hyatt wrote:

“Over the years, I’ve tried lots of different apps for this—Evernote, Notion, Apple Notes. While each had its strengths, they all required time-consuming organization. What I wanted was something that worked intuitively, like my mind does. Enter MyMind.”

I can certainly relate.

I’ve accumulated more than 16,000 notes in Evernote over the years, mostly articles clipped from the internet using Evernote’s web clipper. It had the first web clipper I was aware of and did an exceptional job—although less exceptional as of late—of saving the relevant information and leaving most of the ads and other detritus behind. I considered the web clipper to be Evernote’s most useful and powerful feature.

I also have notes clipped from the web to Notion. To UpNote. More recently to GoodLinks. Things that “resonated” (to use a now popular but ill-defined term) when I encountered them but are scattered in multiple apps and of questionable retrievability. I’ve clipped articles to be read later to InstaPaper, to Pocket, and more recently to Readwise’s Reader. They mostly go unread. Out of sight, out of mind. Multiple junk drawers. My digital garden now a digital weed patch.

Yes, I know. I have a problem. Nevertheless, I still feel a need for a tool—hopefully one tool—that will allow me to quickly capture something I encounter such that it will present itself when it becomes appropriate to engage with it.

Thus, I signed up for a month-to-month “subscription” to mymind; monthly so that I can bail if it isn’t right for me.

Mymind has been discussed in these forums and the consensus seems to be that it is pretty albeit expensive. After using it for nearly three months I find that it fails on three counts.

1. Retrieval is via search but I don’t find it reliable. A card saved recently to mymind that references some research by Harvard psychologist Ellen Langer isn’t retrieved when I search for “Ellen Langer”, the only search term that comes to my own mind.
2. Saving is unreliable. Since yesterday afternoon, mymind spins endlessly without saving any article I’ve attempted to save. (Seems to be a mymind issue. They save right away to GoodLinks.)
3. There is no way to create a link to a card in mymind that I can paste into another app and that will take me back to that card in mymind. I inquired with them about that and they confirm that it can’t be done. There is a way to create a very temporary link to a card that you can send to someone else, but that’s of no interest to me.

The creating a link to paste into another app is a really big deal. Many of us anchor our work days around apps such as Noteplan, Obsidian, and Craft. (I’ve been using Capacities as my daily center for nearly a year and a half). When I save something I’ve encountered there’s often a specific topic or project or task I have in mind that it is relevant to and I want to be able to link to that saved article from Capacities (or Todoist, if that is where I want to be able to access it from).

I recently purchased GoodLinks, which I first heard about in these forums. It does an admirable job of saving clean articles and quite reliably. And one can create a Deeplink to an article that can be pasted into another app. But if I used GoodLinks extensively—with many, many notes—I’m afraid I would be back to tediously organizing it with tags; and with over 7,000 of my 16,000 Evernote notes sitting in the Inbox I’m obviously not good at doing that. Also, GoodLinks does not allow me to “create” a note I write. It only uploads articles via its browser extension.

What I really want is an app that will allow me to quickly, reliably, and fluidly save something I encounter, create links to the ones I choose to create links to, then retrieve them via search or the saved link. And what would be really great would be for it to also present other articles I’ve saved that are related to the one I retrieve, much like DEVONthink does.

Why not use DEVONthink? That could be interesting. I recently “upgraded” to DT4, erroneously thinking it was out of Beta. But it is not and I have not spent any time with the DT4 beta. Nor have I ever made much use of the DT web clipper. If it works well… well, then, a DT4 database of all of the articles I save could be the solution. We know DEVONthink is rock solid. DEVONthink was the number one reason why I migrated to the Mac a dozen years ago. And we know it will be around for the long term.

Three weeks ago Tiago Forte mentioned recall.ai, which looks interesting. It does not have a browser extension for Safari, although they tell me that is on their roadmap. Using it now would mean switching to Chrome—or more likely Arc—and that seems complicated. How, for example, would I access my Apple Passwords?

Has anyone here had experience with recall.ai? Will it do what I want? How about DEVONthink 4?

[KVZ]

@Crystalliere

quickly capture something I encounter such that it will present itself when it becomes appropriate to engage with it.

This is the predicate. How do you imagine a tool that “presents” data when it is “appropriate”?

Are you adding tags? Do you assign other metadata?

“Show me what I need when the time comes” is probably the hardest task for selecting, building, or using PKM.

Personally, when I capture text, etc., I assign it to a “ghost” topic – an anchor. When I’m looking for related information to a problem or idea later on I start interrogating my ghosts, and also search keywords.

FWIW, DEVONthink 4 is pretty good at that, and with growing experience with forming prompts will probably get better. Tana is also pretty good at that, but capturing to Tana is still a bit rough outside of the mobile app.

Katie

[DEVONtech_Jim]

@Crystalliere

How about DEVONthink 4?

Every need you’re mentioning sounds perfectly doable in DEVONthink. Have you downloaded it to see if it suits you?

[WayneG]

@Crystalliere

I’ve accumulated more than 16,000 notes in Evernote over the years, mostly articles clipped from the internet using Evernote’s web clipper. . . . I also have notes clipped from the web to Notion. . . . More recently to GoodLinks. Things that “resonated” (to use a now popular but ill-defined term) when I encountered them but are scattered in multiple apps and of questionable retrievability. . . . They mostly go unread. Out of sight, out of mind. Multiple junk drawers. My digital garden now a digital weed patch.

I still feel a need for a tool–hopefully one tool . . .

My digital garden is also a weed patch, and that isn’t a problem. I’m a former Evernote user that has tried Instapaper and Notion and who currently uses GoodLinks. It doesn’t matter how I capture data because I can quickly retrieve it because I put everything (except sensitive info) into Google Workspace.

When I receive an email I can link it to Tasks from within the Gmail.app. Email, Calendar, Docs, Sheets, Keep, etc are designed to be linked. And when I can’t remember “Zac Wingate’s” name I just search for Cafe Grenel using the Cloud Search app and every document, including non-OCR PDFs, containing that string is displayed almost instantly.

Google isn’t a popular choice with a lot of MPUs. That is due in part, IMO, to people conflating their free apps with their paid business apps. But it solves most of the problems I have faced when trying to use the default and/or native apps available, including backups.

And that is without using Gemini Advanced 2.0, or NotebookLM, which is now standard with all GW business standard and above plans.

Introducing Google Cloud Search

[Crystalliere]

I’m intrigued by your “ghost” topics. Would you care to elaborate?

I seldom use tags, nor do I assign metadata. As I mentioned, an ability to create a link to something I’ve saved away is essential. Here’s a screenshot of a recent entry in a Daily Note in Capacities:

The entry is linked to both a “creativity” topic and an “artificial intelligence” topic in Capacities. Topics are just one of many categories of “objects” I have in Capacities that can be linked to: people, meetings, projects, books, ideas, questions, pages, etc. This entry also happens to have a Deeplink to the article saved in GoodLinks and has been tagged To Read. When I open either the creativity topic or the artificial intelligence topic this entry will show up among the backlinks and appear on a local graph view of everything linked to the particular topic. It probably took less than two minutes to record this resource and link it in such a way I can be confident it will appear whenever I look at notes I’ve made regarding creativity or artificial intelligence. And as mentioned, what would be even better would be a DEVONthink-like ability to also show me anything else I’ve saved away, and long forgotten, that is relevant to my needs.

[Crystalliere]

Recall.ai, mymind, read-later apps and junk drawers

Every need you’re mentioning sounds perfectly doable in DEVONthink. Have you downloaded it to see if it suits you?

Yes indeed!

As mentioned I’ve used DEVONthink for more than a decade. (DEVONthink was number one on the list of reasons I migrated to the Mac). And I recently purchased an upgrade to DT4 but have not had a chance to work with it since I have things I need to wrap up in DEVONthink 3.

Writing is often a cathartic experience for me and as I wrote this original post, assuming there may be other hoarders of information out there who have experienced some of the same difficulties I have—and to some extent that may even include the estimable Michael Hyatt—and realized that one thing I would really like would be a tool that would present me with other articles and notes that are related to the particular one I’m looking for, but had long forgotten exist, just as DEVONthink does… well, only then did it occur to me, why the heck not use DEVONthink? I have it. I’m familiar with using it, although I really ought to learn to use it more skillfully. I can create links to anything I put in DEVONthink. It’s always been rock solid. I can rely on the developers being around long term. And who knows what they’ve cooked up for the DT4 version which, it happens, I’ve already paid for. The more I think about it, the more it seems the obvious solution. And the simple solution. And simple is sustainable.

Author Steven Johnson—who helped develop Notebook LM—used DT to write at least a couple of his books and once wrote that a standard blog post length (roughly 500-1,000 words) seemed around the optimal length for DEVONthink’s traditional AI to function well. I’ve always wondered what would happen if I attempted to unleash DT on the 2,000 research papers I have in PDF. Or all the papers I have attached to Zotero. Maybe whatever new AI capabilities they have in DT4 can handle that. One can hope.

[KVZ]

The context where I use ghost notes is Obsidian (this would also work for Capacities, I think). I insert a link to [[Ghosts/Economy]] for example. That note doesn’t physically exist in the database (vault), but it is shown virtually on the graph. Hence, “ghost”.

Katie

[Crystalliere]

Recall.ai, mymind, read-later apps and junk drawers

The context where I use ghost notes is Obsidian (this would also work for Capacities, I think). I insert a link to [[Ghosts/Economy]] for example. That note doesn’t physically exist in the database (vault), but it is shown virtually on the graph. Hence, “ghost”. Katie

Oh I see. It would be grayed out in the Obsidian graph and probably located out toward the periphery. An Interesting idea. I think I could find it useful but probably more in Obsidian than Capacities.

I’ve chosen to use Obsidian as more of a pure research database and keep all of the daily cruft and notes for everything else in Capacities. Nevertheless, many of the topics in Capacities are things I intend to transfer to Obsidian but in a more deliberate and refined form. I think of these Capacities notes as more of a rough draft for Obsidian, perhaps more like what Niklas Luhmann called “fleeting notes”. Interestingly, one can export a Capacities database as a collection of folders in markdown format. The file can then be opened in Obsidian and behaves as it would had it been created in Obsidian, with all of the folders and links and YAML intact.

[nfdksdfkh]

DT4’s AI implementation is, by design, cautious. If you’re looking for a bit of that AI wizardry where put in a vague and ambiguous search term and it’ll magically find just the thing you were looking for, along with 5 more things you didn’t realise you needed til you saw them, DT4 isn’t doing that. Or trying to do that.

For example if you search for “all articles about heart health in older people” it won’t be able to differentiate between your own notes and published articles. It won’t find articles about heart health unless it has the word “heart” directly in it, so articles that only refer to AAA won’t be found. It won’t know that articles the phrase “over 65” and “older people” are the regarded as the same thing.

This functionality is right round the corner I think. Elephas does this on Mac, its just a bit wonky at times, and can feel like a beta. Claude Projects offers something along these lines but is limited to 50 files per project, so you can’t use it for massive note collections. OpenAI’s GPTs function has the same thing, but even more limited, at 20 files.

I would guess in a year or two there will be tools out there, ready for the mainstream, that you can just dump all your notes & files into, and the AI takes care of all elements of both sorting and retrieval. For me that’s the ideal. The only reason I spend time categorising and filing away my notes and documents is to make retrieval easier. If I could use a tool that did the categorising and filing for me, while also improving retrieval, its a no-brainer, I’d use it.

One thing I might try is simply putting notes as .md files in a folder. There are a few AI based file renaming and tagging apps I’ve seen that work in Finder. There’s also a few apps that will offer AI scan, search and retrieval on notes in a folder. It might be that a combination of these two things is all I need.

[Bmosbacker]

I want to revive this thread.

Has anyone used Recall.ai? If so, thoughts?

Also, has anyone compared it to Notebook LM?

I like the premise of Recall.ai, but I’d want to dump 2,000 articles into it.

Any input will be appreciated!

[krocnyc]

Yes, I’m test-driving Recall.ai. It shows promise, but it is not yet fully baked, IMO—it’s very openly a work-in-progress. (Here’s the roadmap.) I’m also test-driving NotebookLM, which is more fully developed. While there’s some overlap between the two, I don’t think that they are interchangeable.

TLDR: Use NotebookLM for serious exploration of a particular topic or area of concern for which you have already built a repository of source material and for which you need a polished tool. Use Recall.ai if you want to collect things to read, watch, or listen to, but are daunted at the prospect of building your own knowledge base and don’t mind working with something that’s still rough around the edges.

Recall.ai bills itself as a “self-organizing knowledge base.” With some UI improvements, Recall.ai could be a turbo-charged read-it-later app. Every piece of content you load into Recall.ai becomes a stand-alone “card.” The app makes every effort to organize those cards for you via automatic tagging. (You can add tags of your own or re-assign a card to a different tag if you want.) For each card there is a separate workspace that contains a Notebook for your own notes, a Chat Window that will provide you with a summary of the card’s content and answer questions; a Quiz that generates test questions about the content; A Connections page that provides outlinks to source information about people, concepts, terminology, etc related to the card’s contents; and a visual graph of those connections. You throw stuff in, it tries to organize it and build connections for you.

NotebookLM is much more clearly a research and thinking tool for things you either have read, need to read, or need to think about in the context of a particular topic or project.

Ideally, you load related documents into one of your NotebookLM notebooks—“sources” in the app’s parlance—and use the app to help you probe them. “Probe” can mean anything from “Give me a map and hold my hand while we traverse the territory” to “I’ve been over this ground many times before and I have some specific questions about the finer points of its geography at coordinates x, y, and z.”

Your notebook has three windows: The leftmost one is for the sources you’ve added; the middle one is your chat window; the rightmost one is the “studio” that will create audio overviews (the famous ai podcasts), video overviews, mind maps, and various kinds of reports based on the source. You can read the complete text of each source in the source window; NotebookLM also creates a capsule summary of the source and provides a list of key topics that you can explore in the chat window.

I don’t know what Recall.ai uses for AI. NotebookLM is driven by Gemini, and I think the quality of NotebookLM’s AI-generated content is better as a result.

[Clarke_Ching]

@krocnyc

I’m test-driving Recall.ai. It shows promise, but it is not yet fully baked, IMO

Same here. Agree. I love the idea, but it’s a bit clunky/buggy just now, and the AI chat feature has always given me poor results.

[Bmosbacker]

That is extraordinarily helpful, thank you!

[krocnyc]

PS - I don’t find either app (Recall.ai and NotebookLM) good places to read what you’ve put there. If you’re looking for a repository for things to read, neither is really suited to that in the way that, say, MyMind is, especially on mobile.

Recall.ai doesn’t parse PDFs into flowable text the way Readwise Reader does, for instance, which pretty much rules out using it on your phone. Add that to the fact that none of the text presented in the app is scalable—including the content the app itself generates—and it’s simply useless on a mobile device.

NotebookLM’s text presentation is barebones. It parses PDFs into plain text before loading them into your notebook, with no attempt to create a nicely-formatted document. It leaves out any images that might be in the document. It parses the text for the AI to process, not for you to read. The original PDF isn’t uploaded into your notebook, so if you do need to refer to it, you’ll need to pull it up in another app.