Hermes Agent + Obsidian: The Complete AI OS Setup

0:00I build an AI operating system to run my

0:02life with Hermes and Obsidian and I

0:05think building a system like that is the

0:07highest leverage action you can take. My

0:10notes and context live inside Obsidian

0:13and Hermes runs the routines. Notebook

0:17helps me to ask questions across

0:19hundreds of sources and plantator helps

0:22me to avoid AI slope in my Obsidian

0:25world. And every day the system creates

0:27a daily note. Every week it turns those

0:29notes into a review and the dashboard

0:31gives me visibility on how my life is

0:34going. And Hermes can actually improve

0:36its memories and skills the more I use

0:38it. So it grows with me and in this

0:40video I'll go through the whole system

0:42and I show how different parts fit

0:44together and I explain how you can start

0:46building one. First this is my AIOS and

0:50this piece gives me visibility into how

0:52my life is going. I am tracking here my

0:55habits and here I have a quotes so that

0:57each morning I can look up and get

1:00motivated. I have two files. I have a

1:02point A. This is how I live right now.

1:04So here is a file. So it's very

1:06personal. I won't go through it but

1:08essentially just gives a context to

1:10Hermes agent about my current state my

1:12current situation. So I'll also have

1:14point B in north tar for me which

1:16defines uh who I am what I'm doing how I

1:19earn money how do I self-realize my day

1:21is going how do I look like? So I have

1:24those two documents here. So an agent

1:26has context and access to each part of

1:28the system. And the reason because you

1:30want to do that is the more personal

1:32your agent is the better. So if you're

1:34having German agent with no context,

1:36it's going to be as generic as someone

1:38else agent. And really your mode is your

1:40context. If your agent runs on a bad

1:42context, your agent is going to perform

1:45worse. And here's my obsidian. It stores

1:47a context about all of the nodes which I

1:49took. And each point here is a node. And

1:52you can see uh different connections

1:53between node. You can actually go and

1:55navigate. And here's example of Hermes

1:58agent video plan which I was outlining.

2:00And it's a memory layer you own. So it's

2:02your notes. It's all of your context. So

2:05I have this folder. Uh it's called art.

2:07Yeah. It contains all of the notes which

2:09I took. So you can see here have notes

2:11folder. It's all categorized. Here I

2:14have my content, video ideas. I have my

2:17different references and everything here

2:19is structured. So that's a memory layer

2:22and I want to build my personal

2:23knowledge system on Hermes agent because

2:25Hermes agent is open source. It has a

2:28plug-in system and I connected with

2:30obsidian so that everything is stored

2:32locally and I can make sure that the

2:34system is durable for years ahead. Yeah,

2:36let's understand what her agent is. So

2:38her agent is a generic harness and the

2:41advantage of that is you can use any

2:44model with that agent. So if I go here

2:47that's my Hermes. So I can pick any type

2:50of model you would like. So this makes

2:52the system durable and not to be tied on

2:54one provider. So for example, I use

2:55cloud code a lot and cloud code is

2:57closed source. If I don't like something

2:59about it, I can't modify it. I can't

3:01really change it. But with Hermes agent

3:03is fully open source. It's transparent.

3:05I know system prompt. I can create my

3:07own plugins. Here are different plugins.

3:10You can modify and extend your Hermes

3:12agent. And the main part of this agendic

3:14system is agents.md. So this is the part

3:17that explains her Asian on how to work

3:19with me. What are the rules they

3:21establish? What are the boundaries? For

3:23example, this folder contain only my

3:25notes. So I can separate and have a

3:27boundaries between me and AI generated

3:29notes and those folder is fully for AI

3:32generated notes. How should Hermes work

3:34with me? There's a hard rules that

3:36Hermes should not be doing. What is my

3:39world structure? because it has like

3:416,000 nodes and Herm is just getting

3:43lost if you this large of a vault. So

3:46you need to guide it to provide

3:47structure of your vault, how to place

3:49your nodes and how to create templates.

3:52And for our AIS, we have a different

3:55layer. So we have a context layer that's

3:56your files. We have a layer of

3:58instructions that tells the model how to

4:01operate with those files with those

4:03nodes. It has model itself which could

4:06be GPT, it could be a cloud model. We

4:09have a a genic harness itself. That's a

4:11Hermes agent. Then we have a skills. And

4:13skills are playbooks that tell Hermes on

4:16how to accomplish different types of

4:18work. So you don't explain yourself over

4:21and over. Those are all the parts of the

4:23system and let's start building the

4:25smallest complete system. So you're

4:27going to go to Obsidian and here you're

4:29going to see a page. You can download it

4:31for your computer. So it's very simple.

4:33So you just go through this process and

4:36now you're going to build something like

4:37that. this life dashboard. So I'm going

4:40ahead and I'm going to create a new

4:42vault. I'm going to call it Hermes OS.

4:44Okay, great. So we have our obsidian

4:47vault set up. So we have our first node

4:50connected. So and let's create a very

4:52simple structure on how we can start

4:55putting context into her into our

4:58folders. So the first thing is this is

5:01the node and you can see it has a name

5:04welcome. And if I open this Hermes OS,

5:07so I'm going to see this file as well

5:08here. It has the same contents as you

5:11can see. And it's just all plain text.

5:13And this is advantage that you don't

5:14need to have any MCP to connect your

5:16Hermes to Obsidian. You can just open it

5:19in inside of this folder and you're

5:21ready to go. Hermes can natively read

5:23those margin files. Another important

5:25piece here is folders. Uh we can create

5:27a new folder to store the context about

5:30oursel. Let's call it notes. And inside

5:32of this folder, we can create a

5:33subfolder. Let's call it daily. So those

5:35are going to be our daily notes which we

5:38will be taking in the morning during our

5:40morning review. And of course as becomes

5:43more complex you might need to create

5:45more folders to get started. But this is

5:47for now the simplest thing you can do.

5:48We don't want to over complicate this.

5:50So another piece to create is agents.

5:54Here we need to rename it because it's

5:56going to be duplicated. So you see it's

5:59being what we can do is we can do like

6:00that. So and it's ready to go. So in

6:03edges.mdb it's our static memory layer

6:06which going to contain all of the

6:07instructions on how to work with us. So

6:09that's the first step. Let's kind of

6:11setting up the structure. Let's move on

6:13to the next step. Let's uh install her

6:16agent. So for the her agent literally uh

6:18you have a command here. So which you

6:20need to run in a terminal. You can do it

6:22like that or which I recommend I

6:24recommend you to go through this her set

6:26up. So it's going to stop a desktop app

6:29for you like the one which I have here.

6:31And if you will start with Hermes agent.

6:34So it's going to look like that. So I'm

6:36going to create a new profile here

6:37called Hermes OS. Now that's good. You

6:40can see that's how it's going to look

6:42like. And now what we want to do here,

6:44we want to add a new project. So this

6:46project is going to be Hermes OS. And

6:49you want to add here a folder, the one

6:51which we just created. So let me pick

6:53that one. Okay. And you're ready to go.

6:56So here you can see this project like

6:58let's set our model. So, I'm using Oppus

7:005. You can also use Charpt if you're

7:03subscribed to those models. And yeah,

7:05you can just literally ask him, hey,

7:07how's it going? Okay, so that's good.

7:08So, now we can tell Hermes that you want

7:11to set up this morning routine that runs

7:14at 6:00 a.m. and prepares us for the

7:17day. Hey, Hermes. So, I want you to set

7:19up a morning routine uh skill which asks

7:22me questions about, okay, what's going

7:24on? What's on my mind? How do I feel?

7:26How I slept? and I want to have a

7:28template for that. So, let's get rolling

7:30with that. Okay. So, it's looking up my

7:32previous skills which because I already

7:34have this one. So, I'm going to tell it

7:36let's start this from scratch and I'm

7:38going to guide it to use this uh

7:40project. So, you don't have to do this

7:42one if you are just starting cuz uh it

7:44won't find your existing skills. So, it

7:47creates a skill called morning routine

7:49and you can see that this skill is

7:51stored in this folder in Hermes OS

7:54folder. There is a folder called skills

7:56and there is skill.mmd file went to this

7:59folder. So profiles her skills and here

8:03is your skill. So you can see that we

8:05just created it. Okay. And let's

8:07actually go through. Okay. Right now I'm

8:08recording video. It's 8:00 a.m. So

8:10feeling good asleep and schedule is kind

8:12of inconsistent. Went to bed at 12:00

8:15and woke up. So we can do better here.

8:16Yeah. Feeling good energy recording a

8:18video. Yeah. So you can go in more

8:20details. So I just want to show the

8:22concept on how you can leverage this. So

8:24right now Hermes going to save this as a

8:27data not. Uh we can say yes let's do

8:29that. So yeah so you can see that we

8:33just saved it right here. So and what we

8:36want to do here is also like add a front

8:38matter. So front m is very useful to

8:41filter and sort your notes. So with

8:43front m her can essentially just query

8:46your obsidian wall like a database.

8:48Okay. And now we just added this front

8:50mother. So the node type is daily node.

8:53We can track the mood in the morning,

8:55energy in the morning, sleep quality,

8:57all of the pieces. So then whenever you

8:59have more data points, it's going to be

9:02very easier for you to see different

9:04patterns. Let's teach you on how to

9:06store this data. We want to make it in

9:09our daily folder. Let's try that and

9:12let's update the skill so that the next

9:13time we won't make the same mistake.

9:15Let's create it uh here in the root

9:17directory. So we can just ask it to

9:19move. It's in the notes daily. So okay.

9:23Yeah, we finally got it. So now this is

9:25a place we're going to be storing our

9:27notes. Very cool. And Chrome jobs allow

9:29you to essentially run the prompts or

9:32those skills autonomously on a schedule.

9:35To make this happen, you can ask,

9:38let's set up a crown job so that we can

9:41run this one at 6:00 a.m. whenever I get

9:43to my computer. It's kind of everything

9:45is ready so they can start my day. And

9:47this chart is already analyze the

9:49context. So you can see those schedule

9:51jobs. Right here are different examples

9:54you can start with. You can do big you

9:57can have topics new digest a lot of cool

9:59ideas. So I'm just showing you I would

10:02create something like custom one. And

10:03you can actually see that her is

10:05self-improving agent so that it learns

10:08from you and it's updating its memory

10:10because I'm making some corrections

10:12there. Now it's kind of learns from

10:13this. Okay cool. and Hermes just created

10:16this crown job and it explains uh well

10:19just to create this Don but what we want

10:21to do is actually to go through this

10:23process whole process of checkin okay

10:25and after a little bit of back and forth

10:27with her so we created this current job

10:29it starts with the creation of the daily

10:32not according to this template and then

10:33it does this morning check in and ask me

10:36four questions so and then we put this

10:38data into the front matter and that's it

10:40and you can modify this however way you

10:43want so honestly up to you there's so

10:45much flexibility and directions you can

10:46go with that. So, tomorrow at 6:00 a.m.

10:49I'm going to have here the chat already

10:51with this context ready for me and you

10:54can also expand it to go through your

10:56emails, do some news reading, maybe some

10:59YouTube reading all of that is possible

11:01and it's really bounded by your

11:02imagination. Okay, so we have this daily

11:05note. Now you can do this exercise after

11:087 days course or the weekends you can

11:10ask it to do a weekly review for you and

11:12in this weekly review you can just

11:14essentially 7 days of your daily notes

11:17find patterns and understand how to plan

11:19your next week. Yeah what I'm going to

11:21show you next is how to build your life

11:23dashboard that I have here I have

11:25connected it to ordering and I connected

11:27to my YouTube I connected to my habits.

11:30uh we're not going to build something

11:31like that because it requires the

11:32iteration. But I'm going to show you the

11:34path towards that. So for that we're

11:37going to use obsidian plugins. So we

11:39need to extend obsidian to enable

11:41display of those kind of dashboards and

11:44I go to here community plugins and to

11:48press exit restrict mode. So and then

11:50you go to browse and there is a plugin

11:52which we need to accomplish this. It's

11:54data core. So data core allows you to

11:57render those kind of dashboards. Let's

12:00enable that and you are good to go.

12:03Yeah. And to start with uh so literally

12:06what I'm going to do is I'm going to

12:08talk to Hermes. So hey Hermes I want to

12:10create my alignment dashboard and I want

12:13to use data core plugin to display the

12:15data and plot the graph. So we're going

12:17to look at uh my ordering. want to

12:20display this data just let's say my

12:23sleep score and my readiness score from

12:26aura ring okay so for oring I have this

12:29common line interface so right now it's

12:32going to pull this data right so it has

12:34it so let's and we're going to create

12:36this graph we're going to create this

12:38plot and this is going to prove that you

12:40can iterate on this data later on okay

12:42so it's creating this JavaScript code

12:44and scripts to display and get this data

12:48so now we have this script script right

12:50here and you have this dboards folder

12:52here. So right now it's not yet reading

12:55it. So it just created this files with

12:58my sleep score which it pulled from a

13:00ring and you can see that it's not there

13:03yet. So it's not displaying it. So let's

13:05get back to Hermes and tell that okay

13:08it's not working. Okay good. So we have

13:09our first plot we have our first

13:11dashboard in Absian. Oh, and it's a

13:14custom JavaScript component and it shows

13:17my sleep score, my readiness. So, and

13:20you can see that the data is accurate

13:21according to the actual daily notes. So,

13:24you can see readiness score 75 here and

13:27readiness score is also 75 here. So,

13:30it's correct and accurate and pulling

13:32this data from this ring which is very

13:34cool. So, you can also track your steps,

13:37you can track your activity levels and

13:39so on. So, it's your imagination. And

13:41you can also put more components to this

13:44to this room like I have here. So you

13:46can also track your training. You can

13:49track any data. I track YouTube. So it's

13:51pulling from the YouTube API. You can

13:53display it right here. You can track

13:55your habits. You can put any citations.

13:57You can put literally anything. And now

13:59you have a way to accomplish this. So

14:01and one thing that's going to be helpful

14:04for you is cloud design. I'm going to

14:06take a screenshot of this dashboard and

14:07I'm going to go to cloud design. So

14:10let's go. So I want to generate three

14:12different prototypes on how my alignment

14:15dashboard might look like. Right now I

14:16just have a scores from ordering but and

14:19tracking my overall life how it might

14:21look like in that case. Okay. And the

14:23advantage of cloud design is the fact

14:24that they have this design system which

14:27has very good taste well relatively good

14:29taste for the alignment dashboard. Okay.

14:31So it's going to ask you a couple of

14:32questions. Yeah. Okay. So you can pull

14:35data from calendar, task manager, habit

14:37tracker, notes I write, finances. Sure,

14:40let's do the workouts as well. Let's

14:42roll with that. So while it's working,

14:44I'm going to switch gears. So I haven't

14:46talked about this agents.md file. Right

14:48now it's empty, but we can make Hermes

14:52more personal by adding some text here,

14:55adding some guidance here. Let's

14:56actually do that. Hey Hermes, I want to

14:58set up my Asians.mmd and I want you to

15:01ask a question that going to help you to

15:03learn about me so that we kind of make

15:06this Hermes OS more personal to us. So I

15:09think you can understand the pattern

15:11right now that you can literally ask the

15:12model to help you accomplish something

15:15and ask questions and it's going to kind

15:17of very intelligently ask those

15:19questions for you. So yeah, so you can

15:22go with that. So that's obsidian. Yes, I

15:25prefer direct and style. Great. And for

15:28some reason, it's not clicking here.

15:29Let's skip that. So, yeah, I just going

15:31to answer those questions. Yeah. So,

15:33yeah. So, what I do, I work in the

15:35startup. I am also very curious about

15:37personal knowledge management, Obsidian,

15:39taking notes. Uh I record videos on

15:42YouTube to help people build their own

15:44AI second brain to accomplish their

15:46goals in their lives with that system.

15:48In terms of tools, I use I use YouTube,

15:51I use Whisper, I use I use Obsidian.

15:54That's mostly it. Well, what slows me

15:56down is misalignment with the agent.

15:59Then in terms of direction, I think it

16:02should be very direct and should be like

16:04cutting straight to the point, avoiding

16:05like any fluff. Also a helpful

16:07background about myself. I am I'm a

16:09physicist. So I have this physicist

16:11mindset thinking from the first

16:13principles that's going to help you to

16:15understand how to work with me. Yeah. So

16:18that's a good first draft. So let's see

16:20what Herm is going to write about us.

16:22And remember that it's iterative

16:24process. So you can ask me 10 more

16:27questions about me and you can make it

16:29even more personal. So in this context

16:32in agents.md is something that you keep

16:35for a long time that it's don't change

16:38every day, right? Your preferred

16:40communication style, the tools you

16:42should use, who you are, it doesn't

16:44change on a daily basis. And if you have

16:46to explain something here, so you can

16:48talk to Hermes, ask her to update your

16:51agents.mm. Okay, good. And meanwhile, we

16:54have our different design for the

16:56alignment dashboards. So, okay. Oh,

16:58that's actually very cool. I like this

17:00one more. This is very cool and

17:02exciting. And what do we have here? So,

17:04we have a today view. We have a goal

17:06view. We have different sources. And we

17:09have a readiness, score, sleep. We have

17:11our goals. we can understand where ours

17:14went in different notes. So that's very

17:16cool. I I love this part. So I love clo

17:19design for that piece. So and then you

17:21can adjust something. You can provide

17:23comments here. Let's say that like I

17:26don't need those flags. So and then clo

17:29can address those comments and you can

17:30iterate here and once you're ready you

17:33can actually export it and hand off to

17:36your hermits. So you can export it as a

17:38project file. So this is alignment diver

17:40prototypes. So you can save it and then

17:43ask Hermes to let's implement this. So

17:47of course because it's going to be

17:49required to pull some data from

17:51different parts right from your calendar

17:53you would need to like go through this

17:54process and connect Hermes to those data

17:56sources as I did with ordering. There

17:58you go. That's your alignment dashboard.

18:00Highly recommend. Another piece which I

18:02want to highlight is how to work with

18:04Hermes agent. How to stay aligned. How

18:07to avoid redoing the work twice. So I

18:09think that's the biggest mistake which

18:10people make when they're just starting.

18:12They're constantly prompting AI without

18:14understanding the direction where they

18:16want to go. So I highly recommend to use

18:19a tool called planetator that's going to

18:21help you to solve this issue. So for

18:23pollutator uh it's open source tool so

18:26it's free so I'm not affiliated. So for

18:29politator essentially the way to use it

18:31like you can tell her I want to open

18:34this alignment dashboard prototypes in

18:37politator and let's have a plan write to

18:40me plan as HTML artifact to create this

18:43alignment dashboard and ask me find

18:45questions okay and there you go finally

18:48it can actually do that right now for me

18:50and you can see here is your plan to

18:54align with her so it has those

18:56components to build let's say and if you

18:58see that maybe you want to add some

19:00comments. Let's say maybe you don't want

19:01this hours breakdown and let's drop

19:04this. So, and let's going to save that

19:06part and you can see that the comment

19:08appears right here. So, and then maybe

19:11it asks some other questions. What are

19:14the five goals to track? Let's say and

19:17we can select this component and we can

19:19say that I use Apple calendar. Let's

19:22pull this data from Apple calendar for

19:24example. I'm just brainstorming right

19:26now. And then you can just send feedback

19:28or you can copy the comments. And if you

19:31send feedback, so you're going to see

19:32that right now it's going to go back to

19:35Hermes. You see this command got

19:38finished. And then you can see that it

19:40got this feedback. So this is the file

19:42and here is the feedback on this

19:45component, right? I told let's drop this

19:47and now agent can understands what's

19:50going on and how to proceed with this

19:52plan. And you can use beyond that. I use

19:55it for doing my designs like a local

19:58alternative to this clo design or as a

20:00alignment for plants and the work which

20:02you want to accomplish and this going to

20:04help you a lot because you don't want to

20:06say oh agent Google men this thing for

20:08me just one sentence of context and

20:10that's it right because the output of

20:12the agent is going to be very much

20:14misaligned with what you actually want

20:17and this process of aligning yourself

20:20with the agent going to increase your

20:22results going to increase your

20:23throughput by asking proper questions.

20:26And one questions which I really love to

20:28ask Hermes is like ask me top 10

20:32questions that's going to reduce

20:34uncertainty the most. I really love this

20:36one because Hermes kind of understands.

20:39Okay. What are the most fuzzy parts of

20:41our plan and how can we reduce in that

20:44plan? Yeah. So that's a politator. There

20:46you go. And one part of the Hermes agent

20:48which is really exciting is

20:50self-improvement loop where Hermes can

20:53create skills, improve skills and memory

20:56based on your live interaction with the

20:58Hermes agent. Here's the information

21:00from the docs that every 10 user

21:03prompts. There is another sub agent

21:05which looks at the whole conversation

21:08with Hermes and classifies whether or

21:10not something should be saved to the

21:12memory. That's the first layer. That's a

21:14memory layer. So then there's also a

21:16skill layer that kind of does similar

21:19thing that if you have some kind of long

21:22prompt where it takes a lot of tool

21:25calls to accomplish in that case it's 10

21:28tool iteration within a single term turn

21:30when you just send a prompt to Hermes

21:32and it kind of work through it. It also

21:34creates a skills for these kind of

21:35situations and you can check your

21:38journey along the way. There is a skill

21:40called journey. It's a command and here

21:43you can see that you start from like

21:45literally nothing and then her over time

21:48start creating those skills and

21:50memories. So that's an example of the

21:53memory that we should be following the

21:55vault authorship contract that it's file

21:58which is stored in that location as

22:00example of the memory. Here is example

22:02of the skill on how to use Mac OS. I've

22:05been talking about the screen recording

22:07how to do it properly and it'll create

22:08that skill for that. So you can see that

22:10this agent really grows with you. So you

22:12can see your kind of progress which is a

22:15unique feature of the Hermes agent. I

22:17haven't seen that in any of the agents.

22:20Okay. And now let's switch gears to

22:22Gemini notebook. This tool helps Hermes

22:25to work with tremendous amount of

22:27context. Literally like 300 videos, 300

22:30sources at once. And you can chat about

22:32the sources within Hermes agent. So here

22:35is my notebook and example of notebook

22:39is my whole YouTube channel. So

22:41essentially I just have my videos loaded

22:44here into the notebook and I can ask

22:47like any questions here about like any

22:49video which I created. Let's say what is

22:52llmicki in my words and then gymn

22:56notebook going to like answer this

22:57question based on the sources with

22:59actual citations from those. So you can

23:02see here I only have like 30 sources.

23:04You can scale this up to like 500

23:06sources which is something that you

23:08can't do within a hergent itself because

23:10there is a problem of context window and

23:13the context window of those models is

23:15like actually finite. So you can see

23:17here that in one of my sessions I have

23:20literally like 200k token window and one

23:23token is approximately like four

23:26characters. So, and you can see that

23:29like very quickly it doesn't really

23:31scale if you want to like chat with 500

23:33YouTube videos. So, and here is actual

23:36like answer here is different layers of

23:39L and Viki and it actually sides me. So,

23:42that's been something very useful. Now,

23:44the question is how do we actually

23:46connect this to Hermes agent? We don't

23:48want to prompt like notebook. I think

23:50we're past this error that human like

23:52prompting something. We can make our

23:54agents prompt the other agent. I think

23:57that's where it goes. And also you

23:59notice that here I can only ask one

24:01question at a time and that's a big

24:02bottleneck. And within this tool which

24:05I'm going to show you can ask up to I

24:07think five questions in parallel. Yeah.

24:09So in this tool called Gemini notebook

24:11CLA. So that's a GitHub project. So in

24:13this one which I actually use. So as you

24:16might think the way to set it up is very

24:18simple. You just go to Hermes agent. You

24:21can start a new chat. You can tell okay

24:23let's um set up notebook lm. So, and

24:26it's going to guide you through the

24:27process. It's actually very simple. So,

24:29and I'm going to show you right now the

24:31process on how to create a new notebook.

24:33Yeah. So, let's take this channel how I

24:35AI and she has 400 videos. So, and you

24:39can actually do it in this way. So, I'm

24:41going to send this link to Hermes. So,

24:45and let's say I want to upload the last

24:47100 videos from this channel and I want

24:50to ask five question and save into my

24:52notebook. So I want to know about trends

24:55about okay what what's going on right

24:57now what are the unique ideas that may

25:00be underexplored. So that's example of

25:02the front and I have um this skill

25:05called notebook lm. So and this skill

25:08teaches how to actually upload a whole

25:11channel to the notebook. Okay good. So

25:14we found the channel. So it has 400

25:16videos and good it seem that we are

25:19getting our first videos and we are

25:21creating this notebook. how I 100 recent

25:25videos and right now we're just doing

25:27this in batch of 15. So here is the

25:31actual notbook. So you can see right now

25:33it's just uploading this in parallel all

25:35of the interviews all of the videos of

25:38her. Okay. So just now we downloaded 100

25:42transcripts through this notebook skill.

25:45You can see all of the sources right

25:47here. And now we are waiting on getting

25:50those uh questions asked and got back to

25:52our obsidian vault. And it's doing all

25:54of that in parallel. So it saves you a

25:56lot of time rather than just asking one

25:59question then waiting because you can't

26:01prompt it while it's answering the

26:03question. So it can do it for you in

26:06parallel. And the limit I believe right

26:08now if you're on $20 plan for nonbookm

26:10is around 500 questions per day which is

26:14quite generous. Okay. And we are

26:16actually back. So we got our five

26:18questions answer. So here we have this

26:21[snorts] kind of h for this YouTube

26:24channel. It has a cited sources. It has

26:26all of the queries which we run. Here is

26:29the example of the answer which you are

26:31getting. And we have exact citations

26:33saved in our obsidian vault. So we can

26:35also track back to the source and here

26:38the whole answer parts from the notebook

26:41and here you can see the site the

26:42sources which we got and it's mirrored

26:45here in notebook. I haven't even touched

26:47that part. So I didn't have to prompt

26:49here anything and it just ask all of the

26:51questions or ask this question from the

26:54term agent. I didn't do anything of

26:57that. So that's a highly valuable

26:59workflow if I want to learn something

27:01some complex space. You can just upload

27:02a whole YouTube channel on that topic

27:05and just ask questions. And what I

27:07really love here is because exhaust

27:09sources are actually YouTube videos and

27:11they're grounded to the actual human

27:13experiences which is much more valuable

27:15than asking just AI about something. So

27:18it's actually real. So when I treat this

27:19notebook as sort of an oracle where clo

27:23can have this big chunk of contacts and

27:26can ask any question on that topic. If

27:28you want to use herization from your

27:30phone. So the next video shows how I set

27:33it up for Discord where I separate my

27:35chat and profiles. So each chat here has

27:38a separate context and you can think of

27:40it as a different sub agent which has a

27:42different prompt and different purpose.

27:44So in each of this one includes

27:46instructions, memory and sessions. So

27:48why is this one