About this video
We talk about creating an AI chatbot in Obsidian using Smart Connections. It's basically a really easy way to "talk to your notes" with whatever AI model you like.
Full transcript
Edited transcript from the original Roboto page
Hey, how's it going? So today I want to talk about this idea that we've been pursuing for quite a while, and this is the idea that all of this knowledge we can put into some sort of database or tool or, you know, documentation system and we can then pull from that database as a collective as our agency to basically keep hold of all the information, knowledge, and grow as we grow as a team.
Um, now there isn't particularly anything out there at the moment that does this in a simplistic way until we basically found this tool which is called Smart Connections. Now there are a couple of extra bits that I wanted from this, which is offline functionality, the ability to use your own API keys with whatever AI tool you get, because as all of these AI tools kind of accelerate, they're getting better every single day. So you want to always be on the front runner. So if ChatGPT or OpenAI gets a brand new tool, you want to make sure that that's the best one and you're using the best price for your money. And then if it ends up being Anthropic, you know, swap over.
So Smart Connections is really cool because first of all, you can see all the different types of AI functionality you can use. You can use any of these. You can even spin up a local AI for the actual tool, but for the sake of today I'm going to be using Claude 3.5 Sonnet.
And now on the left hand side, you can see all this information that we've actually set up to try and help people get on board to Obsidian because it's not the easiest tool to get onboarded with. In fact, I would recommend this YouTube video that's on the left here. I'll make sure that's in the description.
But from this, we want to ask it something. If you can look on the left hand side, you can see this "How do we create a paragraph in Obsidian?" And so what this will then do when you file this off, it will find the context of whatever markdown file it finds appropriate and it will respond with a little tiny snippet that tells me about whatever it is that I'm asking.
Um, so it says, "Based on your notes, to create a paragraph in Obsidian, you need to press Enter twice. Specifically, your notes mention..." and then it gives you the actual context.
Now this is kind of cool, but it's very basic at this point. But what I want to show you is something very, very interesting because we're not even in the same file. And I want to ask, how are we able to handle blurs, Sanity image? So, how do we handle blur Sanity image? And now why is this interesting specifically is because there's not all that much information out there about Sanity image and blurs, but more specifically there's no information about how we do it in house.
Now you can see here, this is actually how we do it in house. We use the Sanity image URL four, we set a width of 100, a height of 100, and we blur it to 300. So there's no real performance cost by taking this image. And then from there we actually use an overflow and we get this blur large and scale 110.
Now because this note exists, and you'll see the difference here, so if I hit "Handling blurs on the web," you'll see that this looks totally different to the version that Smart Connections passed this back. And this is because it's essentially taking all that information from my document and then running it through, in this case, Claude 3.5, and giving us a very interesting, you know, kind of more context-accurate response, which is incredible.
Um, you might be asking essentially how much does all of this cost, and the answer is pretty much nothing because the person that's building this obviously works from donations and upgrading the account. But ultimately, you're bringing your own API key and he's essentially just built the tool to be able to work with Obsidian community plugins.
So what I would advise doing is if you are going to use this tool, go ahead and say hi to Brian, who is one of the creators of the tool, and upgrade and pay for the extra bit to be able to get the latest version. But yeah, it's an incredible tool. I'd highly recommend it. It's been something that we've been trying to do for months and months and months on end, and this, as far as I'm concerned, is the best thing out there to be able to handle offline, online AI functionality and be able to get context from your documentation.
So yeah, hope that helps. Have a great day. Take care. Bye.
Read timestamped source captions
0:03hey how's it going so today I want to
0:03hey how's it going so today I want to talk about this idea that we've been
0:04talk about this idea that we've been
0:04talk about this idea that we've been pursuing for quite a while and this is
0:06pursuing for quite a while and this is
0:06pursuing for quite a while and this is this idea that all of this knowledge we
0:08this idea that all of this knowledge we
0:08this idea that all of this knowledge we can put into some sort of database or
0:10can put into some sort of database or
0:10can put into some sort of database or tool or you know documentation system
0:14tool or you know documentation system
0:14tool or you know documentation system and we can then pull from that database
0:15and we can then pull from that database
0:16and we can then pull from that database as a collective as our agency to
0:17as a collective as our agency to
0:18as a collective as our agency to basically keep hold of all the
0:19basically keep hold of all the
0:19basically keep hold of all the information knowledge and grow as we
0:21information knowledge and grow as we
0:22information knowledge and grow as we grow as a team um now there isn't
0:24grow as a team um now there isn't
0:24grow as a team um now there isn't particularly anything out there at the
0:25particularly anything out there at the
0:25particularly anything out there at the moment that does this in a simplistic
0:28moment that does this in a simplistic
0:28moment that does this in a simplistic way until we basically found this tool
0:30way until we basically found this tool
0:30way until we basically found this tool which is called smart connections now
0:33which is called smart connections now
0:33which is called smart connections now there are a couple of extra bits that I
0:35there are a couple of extra bits that I
0:35there are a couple of extra bits that I wanted from this which is offline
0:37wanted from this which is offline
0:37wanted from this which is offline functionality uh the ability to use your
0:39functionality uh the ability to use your
0:39functionality uh the ability to use your own uh API keys with whatever AI tool
0:42own uh API keys with whatever AI tool
0:42own uh API keys with whatever AI tool you get because as all of these AI tools
0:45you get because as all of these AI tools
0:45you get because as all of these AI tools kind of accelerate they they're getting
0:47kind of accelerate they they're getting
0:47kind of accelerate they they're getting better every single day so you want to
0:48better every single day so you want to
0:49better every single day so you want to always be on the front runner so if chat
0:50always be on the front runner so if chat
0:50always be on the front runner so if chat GPT or open AI gets a brand new tool you
0:53GPT or open AI gets a brand new tool you
0:53GPT or open AI gets a brand new tool you want that to make sure that that's the
0:55want that to make sure that that's the
0:55want that to make sure that that's the best one and you're using the best price
0:57best one and you're using the best price
0:57best one and you're using the best price you know for your money uh and then if
0:59you know for your money uh and then if
0:59you know for your money uh and then if it ends up being anthropic you know swap
1:01it ends up being anthropic you know swap
1:01it ends up being anthropic you know swap over so smart connections is really cool
1:03over so smart connections is really cool
1:04over so smart connections is really cool cuz first of all you can see all the
1:06cuz first of all you can see all the
1:06cuz first of all you can see all the different types of uh AI functionality
1:08different types of uh AI functionality
1:08different types of uh AI functionality you can use you can use any of these you
1:10you can use you can use any of these you
1:10you can use you can use any of these you can even spin up a local uh sorry a
1:13can even spin up a local uh sorry a
1:13can even spin up a local uh sorry a local AI um for the actual tool but for
1:17local AI um for the actual tool but for
1:17local AI um for the actual tool but for the sake of today I'm going to be using
1:18the sake of today I'm going to be using
1:18the sake of today I'm going to be using Claude 3.5 Sonet and now on the left
1:21Claude 3.5 Sonet and now on the left
1:21Claude 3.5 Sonet and now on the left hand side you can see all this
1:23hand side you can see all this
1:23hand side you can see all this information that we've actually set up
1:25information that we've actually set up
1:25information that we've actually set up uh to try and help people get on board
1:26uh to try and help people get on board
1:26uh to try and help people get on board to obsidian because it's not the easiest
1:28to obsidian because it's not the easiest
1:28to obsidian because it's not the easiest tool to get on boarded with in fact I
1:31tool to get on boarded with in fact I
1:31tool to get on boarded with in fact I would recommend this uh YouTube video
1:33would recommend this uh YouTube video
1:33would recommend this uh YouTube video that's on the left here I'll make sure
1:35that's on the left here I'll make sure
1:35that's on the left here I'll make sure that's in the description um but from
1:37that's in the description um but from
1:37that's in the description um but from this we want to say
1:40this we want to say
1:40this we want to say um how do we so if you can look on the
1:42um how do we so if you can look on the
1:42um how do we so if you can look on the left hand side you can see this how do
1:44left hand side you can see this how do
1:44left hand side you can see this how do we
1:46we
1:46we create a
1:49create a
1:49create a paragraph in
1:51paragraph in
1:51paragraph in obsidian and so what this will then do
1:53obsidian and so what this will then do
1:53obsidian and so what this will then do when you file this off it will find the
1:55when you file this off it will find the
1:55when you file this off it will find the context of whatever markdown file it
1:57context of whatever markdown file it
1:57context of whatever markdown file it finds appropriate and it will respond
1:59finds appropriate and it will respond
2:00finds appropriate and it will respond with a little tiny snippet that tells me
2:02with a little tiny snippet that tells me
2:02with a little tiny snippet that tells me about whatever it is that I'm asking um
2:04about whatever it is that I'm asking um
2:04about whatever it is that I'm asking um so it says based on your notes to create
2:06so it says based on your notes to create
2:06so it says based on your notes to create a paragraph in obsidian you need to
2:07a paragraph in obsidian you need to
2:07a paragraph in obsidian you need to press enter twice specifically your
2:09press enter twice specifically your
2:09press enter twice specifically your notes mention and then it gives you the
2:12notes mention and then it gives you the
2:12notes mention and then it gives you the the actual context now this is kind of
2:14the actual context now this is kind of
2:14the actual context now this is kind of cool but it's very basic at this point
2:16cool but it's very basic at this point
2:16cool but it's very basic at this point but what I want to show you is something
2:18but what I want to show you is something
2:18but what I want to show you is something very very interesting because we're not
2:19very very interesting because we're not
2:19very very interesting because we're not even in the same file um and I want to
2:22even in the same file um and I want to
2:22even in the same file um and I want to ask how we are able to handle blurs
2:24ask how we are able to handle blurs
2:24ask how we are able to handle blurs Insanity image so how do we handle blur
2:29Insanity image so how do we handle blur
2:29Insanity image so how do we handle blur Insanity image
2:30Insanity image
2:30Insanity image and now why is this interesting in
2:33and now why is this interesting in
2:33and now why is this interesting in specifically is because there's not all
2:35specifically is because there's not all
2:35specifically is because there's not all that much information out there about
2:37that much information out there about
2:37that much information out there about sanity image and blurs but more
2:40sanity image and blurs but more
2:40sanity image and blurs but more specifically there's no information
2:42specifically there's no information
2:42specifically there's no information about how we do it in house now you can
2:46about how we do it in house now you can
2:46about how we do it in house now you can see here this is actually how we do it
2:47see here this is actually how we do it
2:47see here this is actually how we do it in house we have um we use the sanity
2:50in house we have um we use the sanity
2:50in house we have um we use the sanity image URL 4 we set a width of 100 a
2:53image URL 4 we set a width of 100 a
2:53image URL 4 we set a width of 100 a height of 100 and we blur it to 300 so
2:57height of 100 and we blur it to 300 so
2:57height of 100 and we blur it to 300 so there's no real performance cost by
2:59there's no real performance cost by
2:59there's no real performance cost by taking this image
3:00taking this image
3:00taking this image and then from there we actually use an
3:02and then from there we actually use an
3:02and then from there we actually use an overflow and we get this blur large and
3:04overflow and we get this blur large and
3:04overflow and we get this blur large and scale 110 now because this note exists
3:08scale 110 now because this note exists
3:08scale 110 now because this note exists and you'll see the difference here so if
3:09and you'll see the difference here so if
3:09and you'll see the difference here so if I hit um handling blurs on the web
3:12I hit um handling blurs on the web
3:12I hit um handling blurs on the web you'll see that this looks totally
3:13you'll see that this looks totally
3:13you'll see that this looks totally different to the version that smart
3:15different to the version that smart
3:15different to the version that smart connections passed this back um and this
3:17connections passed this back um and this
3:17connections passed this back um and this is because it's essentially taking all
3:19is because it's essentially taking all
3:19is because it's essentially taking all that information from my document and
3:21that information from my document and
3:21that information from my document and then running it through uh in this case
3:24then running it through uh in this case
3:24then running it through uh in this case uh Claude 3.5 and giving us a a very
3:27uh Claude 3.5 and giving us a a very
3:27uh Claude 3.5 and giving us a a very interesting you know kind of more
3:29interesting you know kind of more
3:29interesting you know kind of more context accurate response which is
3:31context accurate response which is
3:31context accurate response which is incredible um you might be asking
3:33incredible um you might be asking
3:33incredible um you might be asking essentially how much does all of this
3:35essentially how much does all of this
3:35essentially how much does all of this cost and the answer is pretty much
3:37cost and the answer is pretty much
3:37cost and the answer is pretty much nothing because the the person that's
3:39nothing because the the person that's
3:39nothing because the the person that's building this obviously works from
3:41building this obviously works from
3:41building this obviously works from donations and upgrading the account but
3:43donations and upgrading the account but
3:43donations and upgrading the account but ultimately at the end of the day you're
3:44ultimately at the end of the day you're
3:44ultimately at the end of the day you're bringing your own API key and he's
3:48bringing your own API key and he's
3:48bringing your own API key and he's essentially just built the tool to be
3:49essentially just built the tool to be
3:49essentially just built the tool to be able to work with obsidian Community
3:51able to work with obsidian Community
3:51able to work with obsidian Community plugins so what I would advise doing is
3:53plugins so what I would advise doing is
3:53plugins so what I would advise doing is if you are going to use this tool go
3:55if you are going to use this tool go
3:55if you are going to use this tool go ahead and say hi to Brian who is one of
3:57ahead and say hi to Brian who is one of
3:57ahead and say hi to Brian who is one of the creators of the tool and uh upgrade
4:00the creators of the tool and uh upgrade
4:00the creators of the tool and uh upgrade and pay for the the extra bit to be able
4:02and pay for the the extra bit to be able
4:02and pay for the the extra bit to be able to get the latest version um but yeah
4:04to get the latest version um but yeah
4:04to get the latest version um but yeah it's it's an incredible tool I'd highly
4:06it's it's an incredible tool I'd highly
4:06it's it's an incredible tool I'd highly recommend it it's been something that
4:07recommend it it's been something that
4:07recommend it it's been something that we've been trying to do for months and
4:08we've been trying to do for months and
4:08we've been trying to do for months and months and months on end um and this as
4:10months and months on end um and this as
4:11months and months on end um and this as far as I'm concerned is the best thing
4:12far as I'm concerned is the best thing
4:12far as I'm concerned is the best thing out there to be able to handle offline
4:15out there to be able to handle offline
4:15out there to be able to handle offline online uh AI functionality and be able
4:18online uh AI functionality and be able
4:18online uh AI functionality and be able to get context from your documentation
4:20to get context from your documentation
4:21to get context from your documentation so yeah hope that helps have a great day
4:23so yeah hope that helps have a great day
4:23so yeah hope that helps have a great day take care bye
Questions answered
Editorial answers from the original Roboto page.
Does Obsidian have AI?
Yes. Obsidian itself does not ship with AI features, but it has a rich plugin ecosystem with several AI options. Smart Connections is one of the most popular, letting you chat with your notes using whichever model you choose.
Read the original ↗What does the Smart Connections plugin do?
Smart Connections embeds your Obsidian notes and lets you have a conversation with them. It uses retrieval to pull the most relevant notes into context, then sends that context to your chosen AI model so you can ask questions and get answers grounded in your own writing.
Read the original ↗Can I chat with my Obsidian notes?
Yes. With Smart Connections installed, you can open a chat panel inside Obsidian and ask questions about your vault. Answers cite the notes they were drawn from, so you can jump back to the source whenever you want.
Read the original ↗Which AI models work with Smart Connections?
Smart Connections supports OpenAI, Anthropic, Google, and local models through Ollama. You pick the one you want to use and provide the API key in the plugin settings.
Read the original ↗How do I set up an AI chatbot in Obsidian?
Install the Smart Connections plugin from the Obsidian community plugin browser, enable it, then add an API key for whichever AI model you want to use. Once embeddings finish generating, the chat panel becomes available from the sidebar.
Read the original ↗Is Smart Connections free?
The plugin itself is free and open source. The cost depends on the AI model you connect to it. Usage of OpenAI or Anthropic incurs API charges, while running a local model through Ollama has no per-query cost.
Read the original ↗Scope and context
- Video authors are credited by the original Roboto page; individual transcript speakers remain unassigned.