My AI Note-taking workflows

- Just over a year ago, if you had asked me how I felt about AI and note taking, I would've said that there's no way that I wanted AI near my notes.

Fast forward a year later and the world is different and so am I.

So here I am at a point where I heavily rely on AI to take my notes in many different ways of the many ways that I use ai.

Now, I think probably the easiest to start with is speech to text, meaning taking my voice and turning that into plain text that I can store in my obsidian vault.

In this video, I'm gonna show you three different levels that I've implemented this at and my workflows for each one.

Here's the way that I'm currently thinking of AI's place in my notes Nick Milo makes a distinction between note taking, which is the passive more collecting aspect where you're just trying to write down what you heard or read and note making, which is connecting notes, linking them to each other and synthesizing and deriving insights from them.

I think that AI naturally fits better for me at least in the note-taking aspect.

That way I am kind of outsourcing a lot of the tedium of taking down notes in the first place and not so much the synthesizing and creating new meaning from my notes.

The first tier is completely free, so no credit card information required for either the apps or the underlying LLMs that those apps are using.

I think the best way to start is to just use your phone.

So both iOS and Android have good alternatives for this.

iOS already transcribes voice recordings on the Voice memos app and there is an alternative for Samsung devices as well.

At least if you have one of the later Galaxy devices, then you're gonna be able to use the voice recorder app to do the same thing.

The one that I actually use the most though is something called Spoken.

It is a free app and I'm looking at it right now.

When you start the app, you can look at the dictation models here.

If you see down here I, I'm using a free account, you can upgrade to pro if you want to use some of the frontier models, but I'm not using it for that.

Instead I'm using this local model, it's called Nvidia Parakeet and so you can choose between a lot of different ones.

Apparently there's a newer version of the one that I downloaded that is supposedly faster.

There's also some Quinn ones, so these are free what's called open weight local models that you can download and you, and they're the ones that do the actual transcription.

So how this works is you go into any app, let's say obsidian, but you know it doesn't actually have to be obsidian, it could be like in your browser.

So you could dictate an email.

For instance, I'm gonna use obsidian here and I'm just going to create a new note and all I have to do and you can change what key you press, I'm pressing down the command key and then I can just start typing.

You can see that there's something at the bottom there from spoken that notices that I'm holding this down and that I'm speaking into obsidian and then I'm gonna let go of the right command key and there it is.

Everything I've just said is now text in obsidian.

What I like about this is it's really quick.

There's no data that's being sent on the internet, so this is gonna work.

Even if you have no internet connection and you just hold down a command key that it's pretty easy to trigger it.

You can also change it so that it's more of a toggle thing, so you press once to start the recording and then you press again to stop it.

But I find myself using Spoken Lee more for like quick sort of thing, so I just like to hold it down anyway.

Now, this free tier is what I would recommend if you kind of just want to try it out and see if this will make it into your workflows.

This is I think, the most accessible one because you don't have to pay any money.

What it's not so good for is longer content.

You know, you can do the toggle thing like I mentioned, but there is no asking AI about what you've just recorded.

There's really just translating from words that you, you speak to, words that you've written and that's it.

Sometimes that's enough, but if it's not, then we go into tier two, which is the paid tier.

In tier two, you're basically paying for the convenience layer.

So in tier two you have these paid apps which often have a free component, but I've found that it's really more realistic to think of it as paid because the free tiers are not that generous.

These are tools that you might already be familiar with if you're having meetings online.

So Google Meet for example has Gemini that can record and transcribe what is being said.

Zoom has their AI companion as well.

Auto AI is one that's been out for a while, but the ones that I use, there are two of them are voice notes and granola.

Now granola is really good for meetings and I really like that here in granola.

It brings in my calendar, so I've linked it to my Google calendar so it's aware of the meetings that come up and whenever I start one of them, a little notification comes up in the corner that says, do you wanna take notes?

And it's very easy to just say yes, take notes.

This is actually the summary that was generated from a video I watched by a YouTuber called Book Born.

My partner and I are wanting to read the latest engine crawler Carl book, and we've already forgotten the previous seven books.

So I watched her recap of it and this is the notes that granola automatically took of it.

A really cool thing here is that with granola you can ask questions live.

So I obviously did this already and it's not live, but I can still kind of show you what that looks like.

So I'll say, who is Princess Donut?

And I'll enter that and in real time or apparently after the fact it can answer questions.

So now it's saying Princess Donut is Carl Sapien Cat companion and one of the two central protagonists in the book.

In meetings, I actually find this really useful if I get distracted, especially in long meetings and I'm like, can you just recap the last five minutes because I was not paying attention.

Let me flip back to obsidian.

This is my actual obsidian note for it.

This is the video that I was watching.

I created a macro, so now I just have to open up the command pane and then, and I type granola and I have this link granola one, I hit enter and then I can choose which one it was.

So in this case it was Dun crawler Carl.

So now the granola section is added to the bottom of the note.

So I, I set all of this up in the macro and now that summary is here as well.

So every time I open the obsidian note, the entire summary is also there.

And then at the bottom here, you can also look at exactly what was being said there and it's like nicely formatted too.

So this is using two things.

One is the granola plugin, it's called, yeah, granola sync.

It is a community plugin.

And then it's also using a specific script that I wrote just to be able to link granola transcripts and summaries to the notes, to the corresponding notes in obsidian.

And you can ask your AI to write yourself a script that works for you.

If you want a copy of mine in particular, I do have it in my Patreon vault, so it is also here, granola.

The other thing that I use is voice notes.

Now I tend to use voice notes actually mostly on my phone.

My use case of voice notes is actually a little too personal to show because it's more for like diary style entry.

So I can't show you the interface, but you can also ask the AI there for certain things that span different notes.

So how many times have I talked about this or, or when was the last time that I talked about this person?

So that kind of thing. That requires multiple notes.

It's pretty good at retrieving and on obsidian there's also a voice note sync app that works really well.

And just like you saw with granola, I do also have a macro that runs a script for linking voice notes to my daily notes because my use case of it is more like a journal.

So I want it always embedded in my daily notes.

Some trade-offs on this tier is, well the cost and also the fact that you are losing a little bit in privacy because you are now sending those notes to other people to store on their servers.

There's also the risk of that whole workflow getting interrupted if that that subscription service is no longer functional.

So you do take a little bit more risk with it.

What you gain though is a lot more convenience.

I find that I do voice notes and granola actually more than I use openly, and I think it's just because it's, it's smarter.

So I think that if you're someone who has something specific in mind to capture audio for like a talk at a conference or a video or a voice memo and you're okay with starting and stopping it, then I think the tools in this section are for you.

If you want even more, then there's like tier three, which is the overkill, completely ridiculous level that I'm gonna show you a workflow for anyway, because I am that ridiculous.

Let me introduce you to plot.

You might not have noticed this on my wrist, but it is a little hardware recorder.

It is something that sits on your wrist here.

It doesn't come with this band.

This is something that I got off market, but it comes with a much uglier band or it comes with a necklace or a pin and I kind of mix it up a little based on how obtrusive or unobtrusive I want it to be.

It is basically a recorder and you can just use it as a recorder.

So you can just have it as something that you wear that records your voice, and then at the end of the day you can transfer all of those recordings to your laptop or your mobile device where you can then use AI to transcribe it or you can optionally send those recordings to their cloud.

So what I do here is I use this plot sync plugin, which I did not create.

It is by Leonard sell, but I found that there were some things that I wanted to change that I wasn't sure that everybody would really want.

So you can download his version and install it.

Or if you want, you can also download my version of it, which is on Nicole Vander Holosync for obsidian.

I'll put a link in the description and it just formats it a little better for this particular use case.

It works for me.

You know, you, I would encourage you to choose the version that works best for you.

So that is the obsidian part of it.

That's how I get these transcripts into obsidian and when they come through, they're already nicely formatted and summarized.

One cool thing that I like about plot that I don't get with granola or voice notes is that plot lets you have templates so they're not always just processed in a certain way.

Sometimes I've recorded like games, for example, and when I'm playing a role playing game, I do not wanna have like next actions and agenda items the way I would in a meeting.

So I have a game one and I have a meeting template and then I choose the more relevant template for what I'm recording.

But one thing that I've actually been experimenting with is just using it for life logging.

So turning it on at the beginning of the day and letting it just record everything because PLA devices, or at least this PLA note pin, there are a few kinds of devices including like a card shaped one, but they can record for like up to 23 hours or more I think.

So I've also been experimenting with using it as a life logging tool.

I really love to travel and my partner and I are really enjoying seeing new parts of Portugal.

And so with his consent, obviously I have recorded the entire day of our conversations and those conversations I then feed to my open claw agent.

Now, I'm not gonna get into open claw here, but basically it is, think of it as an AI assistant with a persistent memory.

So this is something that knows you and has an extent, a much more extensive memory of what you like and the things that you're interested in.

And it's also a gentech, so it can do things like create notes.

So I'm gonna show you just briefly how that translates in my obsidian vault.

This is my daily note for May 2nd, and you can see that I have a section here for plot.

That's because I do have macro for that as well.

And it is this one, this linking plot to the daily note.

Each individual one looks like this.

So that this one is I, I was using a template where it's really just a transcript and no summaries and it has these timestamps here for the raw transcript.

Now my open claw agent Iris, and if you wanna know why I call her Iris, then check out this video up at the top where I talk a little bit more about that setup.

Looks through all of the plot transcripts.

So in this case it was 15 hours of recording time or or less 'cause it's five hours per recording.

And she created my open class assistant, created a daily summary of it.

So talks about where we were, the RA coast and a restaurant that we went to.

She basically took my voice transcripts and turned it into a summary of the day.

These are things that I now don't remember, like do I remember that it was called the mesino?

No. All right. This is not without its cons.

First, it has to be addressed.

There are real ethical concerns here.

I try not to record people, especially if they say that they don't wanna be recorded.

That's no questions. I will stop that for you.

But the problem is that there are all sorts of gray areas where, you know what if I'm in a conference and I'm speaking to someone and that person agrees to be recorded, but another person just suddenly comes up and starts talking, that third person is now being recorded.

So do I then have an obligation to stop and say like, Hey, we're recording this.

Is that still okay?

That can be unwieldy in a conference setting, for example, which is actually what I got it for so that I can remember who said what without actually stopping and taking notes.

So I'm still trying to figure out what, where I am on that spectrum and you can figure that out for yourself if, if you want that too.

One of the benefits is that I no longer have to stop at the end of the day and think, where did I go?

What was the name of that restaurant?

Now it just is kind of passively recorded.

At the end of the day, I still have a not perfect but way better record of what I've done than what I normally have time for.

So regardless of which of the three levels you choose or if you choose any at all, what I am getting out of this is that I think voice is now the least friction capture method that exists for me because I get to just do the thing and live my life and have it passively recorded and my tasks waiting for me when I get home back to my obsidian vault.

The real overkill thing here is that I still use all three levels.

I use spoken for just writing an email or dictating something into obsidian.

I use voice notes for more personal diary type entries, granola for meetings, videos and games.

And I use plot as my kind of daily life lager and capture of things that I didn't think to capture.

The result of implementing these tools is that for me, the friction between thinking of something and writing it down is basically gone for me and I've always been a note taker, but I still didn't take notes on everything.

Now it's still not everything, but it's pretty damn close.

And that means that if I can outsource that part of it, then I get to focus on being present, staying in the moment, and if I choose to take notes, it's not going to be what I heard or what happened or anything surface level like that.

I can focus on the actual synthesis.

So I'm getting to that connecting, synthesizing part earlier on in the cycle.

If you want any more details on any of these workflows, then drop something in the comment below so that I know that I should make a video on that.