AI Tools for Obsidian — What I Actually Use (And What I Stopped)

This week, my Reddit intelligence system flagged a survey. Should we make this community AI free or AI limited? As

someone who's been developing an AI tools plugin and testing the nuances of what works and what doesn't, I can relate because after weeks of building and using these tools, I'm noticing

something interesting. Some AI features

something interesting. Some AI features are genuinely useful, and some just take the fun out of working with your notes.

Today I want to show you what I've learned about where AI actually helps and where it starts to feel like you're just watching a machine do your thinking for you.

It's currently been 3 days since Kush posted should we make this community AI free or AI limited. So there's been over 2 and a half thousand votes. The winner

at the moment is allow AI post but require a dedicated tag and filter. And

that's what I voted for as well. So the

concern from Kush is lately there's been a noticeable increase in posts focused on AI workflows, AI vault automations and optimized Obsidian with AI discussions. And he's highlighted that

discussions. And he's highlighted that some members love this direction while others feel it's drowning out conversations about writing note-taking systems, PKM philosophy plugins, and

pure Obsidian geekery. I'll give you some insights on what I've been working on in the AI space with Obsidian in this video and then towards the end I'll give

you the pros and cons of using AI tools based on what I have experienced.

So here I am inside of Obsidian and on the right hand side I've got the AI tools community plug-in open and I want to show you the commands that I'm actually using every day. So, I've

created this note called Claude code reference and it lives inside of my toolkit and this is just a reminder note of what I've created. Using all these AI

tools is great, but if you don't remember what you've actually created, then you're not going to get much use out of them. So, I'll run you through the fix commands first. So, you can see

I've got five fixed commands here. Tool

book, concept, idea, and YouTube videos.

So, we'll navigate up to my personal development. Open my tools and I'm going

development. Open my tools and I'm going to open up a random note here called banish worries. So, we can see we're

banish worries. So, we can see we're missing some YAML properties up the top here. My formatting is a little bit off

here. My formatting is a little bit off with the quote. And I've also missing some when information. So, I can use my fix tool command here, which is this one

here. So, I come over to my AI tools

here. So, I come over to my AI tools plugin for slashfix select tool. Then I need to just select

select tool. Then I need to just select banish worries as the note that I want to run this command in. Send that

command to my AI tools claude code. It's

going to access the skill that I've created for fixing this note. So I'm

going to allow that. It needs to know about my tools template. So it's going to compare it against my tools template.

So now I've got a change to review here.

So it wants to add some YAML properties.

So, we can see we've got the date created, some time information, as well as some tags, a link to an idea and topics. We are missing summary and

topics. We are missing summary and source, but that's okay. Got some more time information here. It's going to keep my action item, but put it in a better formatted quote. And we've still

got our what, when, and how. So, we can allow that. And now I've got a nice

allow that. And now I've got a nice cleaned up note. So, I can see my properties here. So, I can take out

properties here. So, I can take out unfinished. Now I need to probably add

unfinished. Now I need to probably add category, summary, and source. So what I can do there is I can use another command called review note. So I'll

select review note. Make sure I've got the note selected. Send that command to my AI tools claude code. It's going to access that skill. I'm going to say allow.

So after a while, we get a response back. So we can see that it's promoted

back. So we can see that it's promoted it to a fern because it's between 100 and 200 words. Then we've got some critical improvements that could be made. So we've got an empty when

made. So we've got an empty when section. I know about that one. We're

section. I know about that one. We're

missing the source from our YAML property here. Some quality improvements

property here. Some quality improvements that we should consider fixing. So add a one to two sentence of the summary.

Missing wiki links in action bias. So

the what section mentions action bias but doesn't link to action bias. So

we've got action bias here but it's not linking there. There's no related front

linking there. There's no related front matter to related tools. So, we've got a few tools here that we could link it to.

And then we've also got some enhancements worth considering. So, the

what section only has two sentences could explain what more context and why action bias matters for banishing worries. Some duplicate content in the

worries. Some duplicate content in the fourstep process and empty alias field.

So, we'll just say please fix all those recommendations. So, then it'll draw up

recommendations. So, then it'll draw up a plan. So, we've got our first change

a plan. So, we've got our first change here. We're adding a category, mental

here. We're adding a category, mental health, a summary, a source, and aliases. So, I'm going to allow that

aliases. So, I'm going to allow that change. I can scroll up and see that in

change. I can scroll up and see that in real time here. So, we can see everything being ticked off in our list here. So, we've got our next change to

here. So, we've got our next change to approve. So, we're going to change our

approve. So, we're going to change our what information. Going to add some when

what information. Going to add some when information. And we're also going to put

information. And we're also going to put a tip. So, we're going to allow that. So

a tip. So, we're going to allow that. So

now we have our what linking to action bias. I can hold the control and go have

bias. I can hold the control and go have a look at that note. Now we have a few dot points for when. So when to use this tool and we've got a nice little

four-step call out for our how. And then

we get a summary from claude code. So I

quite like having a note already using the fix command and then following up with the review note. And I find that really cleans up the note and makes it

more useful to me. and it's more collaborative. So, I'm working with the

collaborative. So, I'm working with the AI instead of letting the AI do absolutely everything. And then I can go

absolutely everything. And then I can go back and forth and further enhance this note or ask for an opinion on how to improve something. So, if I'm not happy

improve something. So, if I'm not happy with this quote, we can do some reflection down the bottom here. Let's

have a look at another example. I'm

going to go up to the properties here and grab one of these related notes. So,

we'll open this decision action. So we

can see this one is very similar, but instead of a tool, this time it's an idea note that needs fixing. So we can see it's selected in my AI tools here.

So I'm just going to go fix and select idea. Go ahead and send that to Claude

idea. Go ahead and send that to Claude code. It's going to go access my fix

code. It's going to go access my fix idea skill. So I'm going to allow that

idea skill. So I'm going to allow that skill to be read. All right. So now we have our note. So we've got our YAML properties here. So, we're still keeping

properties here. So, we're still keeping mindset, mindfulness, got the date created. We're missing a topic and

created. We're missing a topic and summary, but we've got a bit of other information there. Still got our quote

information there. Still got our quote as well as our reflection and our external sources down the bottom. So,

I'm going to select a layer. So, that

cleans up my notes a little bit. It

didn't link my meta details here, but that's fine. And we're still missing

that's fine. And we're still missing topics up the top here. We get a nice little changes made here. So I might want to go back and say please add links to meta details.

So we've got our before and after. So

we'll hit allow. Now we have our links on our meta details. All right. So I can run my review note command again. So

we'll go review note. Select this note.

Send the message. So we'll now go read my review note command which is a skill.

So, we've got the note is in good shape overall. The main gaps are missing wiki

overall. The main gaps are missing wiki links from action bias and empty front matter fields. Please fix those. So,

matter fields. Please fix those. So,

it's going to fix this empty topics hopefully. And also put in an alias.

hopefully. And also put in an alias.

So, we've got our plan here. Add

aliases, topics, and related to front matter. So, we've got our little change

matter. So, we've got our little change here. So, now I got two aliases and

here. So, now I got two aliases and topics and related. So I'll hit allow.

We can go up and have a look at that in real time. So we've got three notes here

real time. So we've got three notes here related three topics now. And we also have an alias up here. So we don't like one of those. We can take them out or we can just leave them there. So now we're

going to fix redundant fourstep process content which is this content here. So

we'll allow that one. So now we have everything complete. So we can see it's

everything complete. So we can see it's just put dot points there. I actually

should have put numbers, but that's fine. We get a breakdown of what's been

fine. We get a breakdown of what's been fixed.

So, let me show you the workflow I've been using this week when researching Agentic AI. So, on YouTube, I found this

Agentic AI. So, on YouTube, I found this video explaining Agentic AI, the good, the bad, and the ugly by explaining computers. So, usually what I'll do is

computers. So, usually what I'll do is I'll click the video, then I'll use my get recall extension, and that gives me a concise summary of the video. So, you

can see it breaks it down into each section. So, we've got Agentic AI

section. So, we've got Agentic AI basics, the good, the bad, and the ugly.

So, I can either jump to parts of the video by selecting the timestamps, and then just quickly scrub through to find if it's going to be something useful. I

quite like this video because it's got these cards that lay out all the information for you. So from there, I'll just go up to my web clipper. It'll

detect YouTube and load the YouTube template. All my YAML properties are

template. All my YAML properties are predefined. All the body of the note is

predefined. All the body of the note is also predefined and it places it in my vault YouTube folder. I just need to choose my model here. So all I need to

do is hit add to Obsidian. After a few seconds, it will appear inside of Obsidian. So you can see it's brought in

Obsidian. So you can see it's brought in the title here and we've got some predefined YAML properties. So I don't need to fill out any of this. The web

clipper does it for me. Under meta

details, I have some date information.

Then I've got a brief summary of the video, some key takeaways, a summary inside of a call out, the description of the video, then a link that links to a

community plug-in called media extended, and then just a little Metabine embed here for some buttons to control this plugin. So when I select the link there,

plugin. So when I select the link there, it loads up in the right pane, and then I can hit play and scrub forward until I find the part of the video that I want to take notes on. So I could either

write it down, but if I'm struggling for time, what I can do is I can use power toys which has a tool called text extractor and this is on Windows and the

shortcut is Windows key shift T and then I can just grab all the text there. Come

over to my note and paste it in. And

that just saves me typing out word for word. I want to do it in my own words

word. I want to do it in my own words then I can do that later. I just want to get the information captured into the note. So there we have there. And then I

note. So there we have there. And then I just simply put a title or a heading three. What is Gentic AI? And I would

three. What is Gentic AI? And I would just continue through the video taking all my notes. Once I've got my notes, I can close my video. And if we scroll down, we can see that I've done that

below here. So I've got what is Agentic

below here. So I've got what is Agentic AI? What are AI agents? We've got what

AI? What are AI agents? We've got what are AI agents from Google? And we've got AI orchestration notes, communication standards. So we can start branching off

standards. So we can start branching off from this note to link notes. So you can see I've done that here with the modal context protocol. So I've open up that

context protocol. So I've open up that one. This one has been created as a

one. This one has been created as a concept and it's talking about the modal concept protocol in there. So if you find that you want to make a note on agent to agent, then you can do that one

there. And there's a note on agent to

there. And there's a note on agent to agent. So you can see my notes in there.

agent. So you can see my notes in there.

And we can do the same for Microsoft autogen. I see I haven't created a note

autogen. I see I haven't created a note for that one yet. So, I could go make that one now. And then what I tend to do from there is I won't leave my notes in there. I'll just grab them all like

there. I'll just grab them all like this. And to save duplicates, I

this. And to save duplicates, I recommend cutting, but if you want to have two copies, it doesn't really matter. So, you can copy it. And then I

matter. So, you can copy it. And then I use my quick add command here. Run quick

add new concept. Just call this one aentic AI. And then I just paste in my

aentic AI. And then I just paste in my notes. So now I've moved my notes from

notes. So now I've moved my notes from this YouTube video to this note here and close that one. I could put the source in there if I wanted to. So I could put

my related here so it links back from here. I could use my AI tools to do a

here. I could use my AI tools to do a back and forth approach to fix up this note if I wanted to further improve it.

So I could load in my AI tools into the right pane here. And I've got my commands that I could run. So I could go fix concept select this note that's

going to fix this note. So send that command. So I need to allow it to do

command. So I need to allow it to do that. So I say allow.

that. So I say allow.

So after some thinking time code has made some fixes to my agentic AI note.

So because it's 400 plus words, it's taken out the seedling and upgraded it to a tree. So you can see that's been applied here. We've now got a category.

applied here. We've now got a category.

So we can see we've got technology. Also

got a topic area which is artificial intelligence. The meta details already

intelligence. The meta details already present. So it's skip that one. And it's

present. So it's skip that one. And it's

already got proper YAML. So it's skip that one as well. If we want to review more, we can see what was changed here.

And there's also another summary down here. So it's all automatic. It does

here. So it's all automatic. It does

take a bit of time. So it really depends. Like I think there is enjoyment

depends. Like I think there is enjoyment in manually making these changes versus letting the AI automatically do it for you. And that goes for these

you. And that goes for these communication standards here. If I don't have notes for these here, then yes, I could write a skill to go create those.

But sometimes it's just more enjoyable to put links on what you're interested in and then go make the note yourself and then go back and forth with AI tools

in order to further improve that note.

If I open up the left pane and open my toolkit again, I'm going to go back to my Claude code reference here. We'll

just clear this tool here. Now, I want to show you how you can research with an agent.

Let's now have a look at an enentic AI research assistant scenario. I find

Obsidian AI tools good for basic skills and commands. So, at the moment, it's

and commands. So, at the moment, it's using Claude code and Gemini using the agent client protocol, which is a node package manager that lives on your

computer. and that connects to your

computer. and that connects to your Obsidian AI tools subscription which is $7 per month at the time of making this video. There are some limitations to the

video. There are some limitations to the Obsidian AI tools plugin. For example,

when we step into Aentic AI, in other words, running an orchestrator agent with multiple sub agents. In theory, it

can plan and execute similar tasks, but we're going to get better results using a claude subscription with Anthropic. I

would recommend a pro plan of $20 per month. You don't need the max plan. And

month. You don't need the max plan. And

this will allow you to run complex commands and spin up multiple agents.

Let me show you how we can run an Aentic AI research assistant agent using a claude subscription. So to run an Aentic

claude subscription. So to run an Aentic AI research assistant, I need to run my research deep command, which is this second one here. So if we jump into VS

Code, you can see that I've got my research deep markdown file in my commands folder inside of my claude

folder which lives inside of my Obsidian vault. Let's have a look at the command.

vault. Let's have a look at the command.

So the argument hint is topic comprehensive parallel research spawn sub agents writes note to bold. So this

is a task and then we've got a comprehensive research using the research agent sub agent. Use the task tool to invoke the research agent sub

agent with the prompt research this topic comprehensively and then the agent will spawn four parallel web research sub aents compile all the findings into

a structured note and then it's going to write it to this location here. So it's

my research notes inside of my spaces folder. So in order to do this, it's

folder. So in order to do this, it's going to use my agent here, research agent. We can see you are a universal

agent. We can see you are a universal research orchestrator sub agent. Specify

a research topic. Extract the topic from your prompt and execute the workflow below immediately. No confirmations

below immediately. No confirmations needed. And then we've got the sub

needed. And then we've got the sub agents here and all of their tasks. Then

we've got the next step which is to capture the date information. Step

three, write the research note, place it into a location in my vault. Then we've

got the markdown template so it knows exactly how to format the note. So we

can either run it from VS Code with Claude code on the right here. And this

is using the Claude Pro subscription, so not the AI tools subscription. So I can use my command research deep but I need to choose a topic. If I don't put a

topic it will detect that I haven't specified one. So let's just say agentic

specified one. So let's just say agentic AI. So now it's going to go ahead and

AI. So now it's going to go ahead and have a look at my research agent here which is this agent that we just went through. And now all four sub agents are

through. And now all four sub agents are running in parallel. So we've got these sub aents here. We got our fundamentals tools and ecosystem criticisms and

limitations. and then examples and use

limitations. and then examples and use cases. So we'll just approve these

cases. So we'll just approve these queries and these are just web search queries that claude code is going out to perform those tasks. One of the biggest

challenges of using automation AI tools is waiting for them to complete. So what

I like to do is I actually like to work in Obsidian while I'm waiting. Sometimes

I'll use either VS Code on a second monitor or I'll just use the terminal and have interactions with Claude via the terminal. All right, so we can see

the terminal. All right, so we can see we've got a new note now inside our research notes folder called Agentic AI research.md.

research.md.

So let's go have a look at what that looks like. Let's jump into Obsidian. Go

looks like. Let's jump into Obsidian. Go

to our research notes and we can see our new note is here. Aentic AI research created today on the 20th of February.

So, we'll open that one up. We've got a button that's not working here, but we do have some YAML properties which are looking good. I can see that they don't

looking good. I can see that they don't match my template exactly, so I'd have to do a little bit of work there. Date

information has come in. Okay, got a little call out to tell me it's been autogenerated. So that's kind of handy

autogenerated. So that's kind of handy if I was doing a random note look up or I stumbled on this note in the future.

It is nice to know that it was autogenerated. So we've got some Aentic

autogenerated. So we've got some Aentic AI research fundamentals, how it works, got some tools and ecosystem criticisms

and limitations, examples and use cases and then all the sources that it used.

Could be a good research note. It's got

research there. So, that's another way to collect research notes. Personally, I

don't like to do it this way. I think it just spits out a whole bunch of text and I've got to go through and find what's useful. Some of it might not be useful.

useful. Some of it might not be useful.

It might be just noise. As good as Aentic AI is, and the automations are great, sometimes they're not really useful. You're just capturing the

useful. You're just capturing the information without actually thinking and engaging with that information. So,

that's how you can use Aentic AI with your Obsidian Vault. Now, you'll see I've got all my fixed commands here.

I've got my fix book command, my fix concept command, my fix idea command, fix tool command, fix YouTube video command. I thought my format note

command. I thought my format note command would be useful. So, this one just scans my note and improves the formatting. Most of the fixed commands

formatting. Most of the fixed commands are doing what I want them to do. The

research commands also I just find it better to do my own research rather than rely on clawed code to go out to the internet. I might not be interested in

internet. I might not be interested in all the information. I just want to capture what resonates with me. There's

also this one here that I made which is create concept. So this one creates

create concept. So this one creates concept notes from AI conversation related topics. So, it'll look for

related topics. So, it'll look for concept topics in the note and then create a concept note based on that. I'm

still experimenting with this one. I

don't really think it's necessary either. I like to just identify what

either. I like to just identify what notes I want to link out from my main note and start making notes on instead of asking Claude Code to do that for me.

I'm still experimenting with that one.

So, I've got a few here that I've also been experimenting for my YouTube channel. So, YouTube formula, research,

channel. So, YouTube formula, research, and script. I ran all three of them, and

and script. I ran all three of them, and it did write me a script for this video.

I just didn't feel like it was authentic. I thought, this is not what I

authentic. I thought, this is not what I want my channel to be. I don't want it to automatically write my content and then I read off a script. It just

doesn't feel authentic to me. I can see a lot of YouTubers using that type of method, but for me, it's not something I'm interested in doing.

To summarize the Obsidian AI tools plugin, I would treat this as a collaborative approach. So, the pros of

collaborative approach. So, the pros of this AI tool is you can stay in control.

Changes are visible and traceable. The

back and forth keeps you engaged and absorbing the process. It's a

collaboration rather than a delegation.

The fixed skills and enhancements are genuinely useful. The cons I've

genuinely useful. The cons I've identified so far is it is a little bit slower than full automation. It requires

active involvement during each step.

It's not suitable if you need bulk processing. It can make you lazy and

processing. It can make you lazy and hijack critical thinking. If we look at claude code as a full automation agentic

approach using our terminal or VS code, the pros are fast output. It can handle complex multi-step tasks. It's powerful

for developers and iterating on code or notes. I'd say the cons are the output

notes. I'd say the cons are the output can arrive faster than you can absorb it. So notes become hard to recognize as

it. So notes become hard to recognize as your own unless you flag them with a tag or call out. The trust gap between your

content and AI additions becomes wider and also the fun of the process disappears because it's hard to audit what's actually changed.

So I find everyone is talking about agentic AI at the moment. autonomous

agents, AI doing everything for you in Obsidian. On paper, it sounds great. It

Obsidian. On paper, it sounds great. It

saves time. It automates the boring stuff and you can get more done.

However, if the output is faster and you don't remember what you've created, then what's the point of having those notes anyway? So, I encourage you to use these

anyway? So, I encourage you to use these tools and experiment with them, but don't rely on them 100% to do everything for you. Part of the enjoyment of

for you. Part of the enjoyment of Obsidian is engaging with your notes and thinking and writing as you go, not letting AI do everything for you. The

post on Reddit this week is valid. My

stance is AI is not going away. We need

to embrace these tools, but we don't want to fully rely on replacing our thinking or note making. So, that's

where I'm at with AI tools right now.

still building, still testing, still figuring out what works. If you are using AI tools in Obsidian or thinking about it, I'd love to hear your take.

What features do you actually find useful? Where do you draw the line? Let

useful? Where do you draw the line? Let

me know in the comments. And if you want to try AI tools or yourself, the link is in the description. Thanks for watching and I'll see you in the next video.