Build An AI Second Brain Knowledge Base (Step-By-Step) — Transcript

Full transcript

1. 0:00So, I just built a second brain
2. 0:01knowledge management system that has an
3. 0:03entire wiki built in that I can chat
4. 0:05with. It will pull any information from
5. 0:07that second brain when I chat. It's got
6. 0:09a built-in CRM. I can journal, and it
7. 0:11will actually look at my wiki knowledge
8. 0:13base and try to help me with whatever
9. 0:15issues I'm going through from my journal
10. 0:18by looking inside of the wiki. And of
11. 0:20course, it's got all of the content that
12. 0:23I've saved from around the web,
13. 0:24including YouTube videos and articles
14. 0:26and tweets and podcasts and just tons of
15. 0:30stuff that I've injected into this that
16. 0:31is all accessible directly from a chat
17. 0:34or from journaling. It is really, really
18. 0:36sweet, and I'm going to break down how
19. 0:38the whole thing works and how you can
20. 0:39build one for yourself right now. Most
21. 0:41second brain systems are just like
22. 0:42storage, right? You dump your YouTube
23. 0:45transcripts and your articles and your
24. 0:47blog posts and your podcasts and just
25. 0:49everything that you're interested in,
26. 0:51you just dump it all into one place.
27. 0:53Problem is, that's kind of where the
28. 0:55information just goes to die. Unless
29. 0:57you're like actively going back through
30. 0:59and reviewing the notes all the time and
31. 1:01searching through your second brain,
32. 1:03it's just a dumping ground for
33. 1:04information that I never go back and
34. 1:06look at later. So, for the knowledge
35. 1:08management system that I'm going to
36. 1:09build, there's three core pillars that I
37. 1:12want to build for mine. Number one is
38. 1:13the wiki/knowledge base. This is where
39. 1:16I'm going to store like everything from
40. 1:17around the web that I find. YouTube
41. 1:19transcripts, articles, podcast
42. 1:21transcripts, tweets, you name it, it all
43. 1:24goes into this wiki knowledge base
44. 1:26section. Number two is my CRM. So,
45. 1:28whenever I go to a events and I meet
46. 1:30people or I jump on Zoom calls with
47. 1:33people, I want to remember those
48. 1:34conversations and I want to be able to
49. 1:36recall them in the future. I also want
50. 1:37to store details about those people, how
51. 1:40I met them, where I met them, some of
52. 1:42the discussions we had, any sort of
53. 1:44contact details I got from them, email,
54. 1:46phone number, address, whatever. They
55. 1:48can all live in this sort of CRM element
56. 1:50of this bigger second brain that I'm
57. 1:52building. And the third element is where
58. 1:53this all gets pulled together, and And
59. 1:55the journal. Now, I'm a big journaler. I
60. 1:57journal pretty much every single day.
61. 1:59When I have good days, I journal about
62. 2:01what went right, the things I'm excited
63. 2:03about, gratitude, that sort of stuff.
64. 2:04When I have rough days, I journal about
65. 2:06the things that are bothering me. My
66. 2:08videos not performing as well as I want
67. 2:10them to, having a creative block and not
68. 2:12knowing what to make videos about. I do
69. 2:13a lot of travel and I debate a lot about
70. 2:16whether the travel's going to be worth
71. 2:17it or not. I journal about pretty much
72. 2:19everything in my business. So, these are
73. 2:21my three ideal inputs. For you, it might
74. 2:24be clients or workouts, research papers,
75. 2:27recipes, sales calls, classroom notes.
76. 2:30The point is, the knowledge base sits at
77. 2:32the center and then everything else sort
78. 2:34of connects to it. The two elements that
79. 2:36I think are probably the most useful to
80. 2:38the most amount of people are going to
81. 2:39be the wiki and the journal. Maybe the
82. 2:41CRM isn't what you need. Again, maybe
83. 2:43it's your classroom notes, your
84. 2:45workouts, your recipes, et cetera. So,
85. 2:47here's a rough drawing of what I have in
86. 2:50mind. So, you've got your knowledge base
87. 2:51that lives at the center of all of this.
88. 2:53All of this knowledge is going to live
89. 2:55in Obsidian. I'll get into the whole
90. 2:57building process in a second. You're
91. 2:58going to save articles from around the
92. 3:00web, YouTube videos, you know, podcast
93. 3:02notes, whatever you find around the web
94. 3:04that's relevant to you, you're going to
95. 3:06save it with a simple Chrome web clipper
96. 3:08and it's going to save into your
97. 3:09knowledge base. The CRM that I just
98. 3:11mentioned, notes about people you met
99. 3:13and where you met them and all that kind
100. 3:15of stuff gets saved to the knowledge
101. 3:16base. Meeting notes, I personally use
102. 3:18Granola to record my meetings and take
103. 3:20notes for me. Those meeting notes can
104. 3:22automatically be injected into the
105. 3:24knowledge base. And then you have
106. 3:25journal entries. This is the layer where
107. 3:27you actually interact with your
108. 3:29knowledge base. You journal on what
109. 3:31you're dealing with right now and
110. 3:32ideally it's going to pull from the
111. 3:34knowledge base that has all of this
112. 3:35other information in it to ground the
113. 3:37responses to your journal entries. This
114. 3:40will make more sense as I go. Please
115. 3:42excuse the PowerPoint style slide here,
116. 3:44but I really want to explain what I'm
117. 3:46trying to build. So, here's the system I
118. 3:47imagine. You save your articles, your
119. 3:49transcripts, et cetera, into this
120. 3:51system. We'll do that using a web
121. 3:53clipper. The AI layer in the background
122. 3:55that we're going to build then
123. 3:56summarizes this stuff for us. So, it's
124. 3:58not just a giant transcript, it's
125. 4:00actually sort of the bullets and just
126. 4:02the information we need to know. The AI
127. 4:04is also going to extract people,
128. 4:05companies, tools, ideas, and themes, and
129. 4:08sort of break those off. That's where it
130. 4:10becomes kind of like a wiki. You could
131. 4:11click into the tools page and it will
132. 4:13list off all the tools that have been
133. 4:15mentioned across everything that we've
134. 4:17saved. You can click into one of the
135. 4:18tools and it will mention where and what
136. 4:20video that came from. And that goes for
137. 4:23all of these little categories here. I
138. 4:25also wanted to auto-link related notes.
139. 4:27So, if I have multiple videos about how
140. 4:29to build something with OpenClaw, they
141. 4:31all get cross-referenced to each other
142. 4:33and I can click around and sort of jump
143. 4:35into others. If you're familiar with
144. 4:37other second brain systems or like the
145. 4:39Zettelkasten system, it's essentially
146. 4:41that same concept of interlinking. I'm
147. 4:43then going to let the journal directly
148. 4:45into the system. So, when I do write
149. 4:47journals, it responds like ChatGPT, but
150. 4:50it's actually grounded in my own saved
151. 4:51knowledge. So, it's not going to just
152. 4:53respond with what ChatGPT would have
153. 4:56responded with. It's going to respond
154. 4:57with something like, "I see you're
155. 4:59struggling with ideas for videos. Well,
156. 5:01you saved this video 3 days ago that
157. 5:03says you should do this, this, and this.
158. 5:05You also saved this video 2 weeks ago
159. 5:07that gave this advice when you're
160. 5:08struggling with video ideas." And it
161. 5:10will actually pull from the knowledge
162. 5:12that I've saved, making it more tailored
163. 5:14to exactly what interested me and what I
164. 5:17found valuable over time as I save stuff
165. 5:20into the system. It's also going to use
166. 5:21AI to find patterns from past journal
167. 5:23entries. So, if I'm constantly
168. 5:25journaling on the same thing over and
169. 5:26over again, the same struggle, it's
170. 5:28going to see that as a pattern and take
171. 5:29that into account when it's responding
172. 5:31to my next journal entries. It should
173. 5:33resurface relevant ideas when I need
174. 5:35them, and it should also let me save
175. 5:37notes about people and connect those
176. 5:39people to ideas, companies, events, and
177. 5:40conversations. So, the CRM and the
178. 5:42meeting notes that I'm constantly
179. 5:44saving, that should also be connected to
180. 5:46ideas and journal entries and all of it
181. 5:48sort of pulled together in this big
182. 5:50mashup of like here's the conversations
183. 5:52I've had, here's what I'm journaling on,
184. 5:53here's what I've saved that I found
185. 5:54interesting, and it's all in this big
186. 5:56old soup of content that I'm saving, but
187. 5:59that big old soup of content that I'm
188. 6:00saving is the grounded information that
189. 6:03is getting pulled when I write my
190. 6:05journal entries and then get a chat
191. 6:06response based on those journal entries.
192. 6:08All right, now that I've completely over
193. 6:10explained this, let's just jump into the
194. 6:12process I'm going through to build this.
195. 6:14>> There's a cool new feature from
196. 6:15Hostinger that's made for deploying AI
197. 6:17agents at home. So, if you've ever
198. 6:19wanted to run something like OpenClaw,
199. 6:21but don't know where to start or find it
200. 6:23really complicated, this is going to
201. 6:24make the process a lot easier. I
202. 6:26actually use OpenClaw myself, so I can
203. 6:28attest that the setup can get pretty
204. 6:30technical. Right now, it's deployed on
205. 6:32my DGX Spark and my favorite thing it
206. 6:35built is simple CRM that when I meet
207. 6:37people at events, I can save quick notes
208. 6:39in Slack and then later ask it to remind
209. 6:42me what we talked about. So, it's great
210. 6:44for automating everyday things like that
211. 6:46to full-on businesses and these huge
212. 6:49agentic systems, which I've also seen
213. 6:51people do. And this new feature from
214. 6:53Hostinger makes it a lot easier to set
215. 6:55up. You can either pick a managed
216. 6:57OpenClaw plan or a VPS option. The
217. 7:00managed version is honestly the easiest.
218. 7:02It's a one-click deployment that
219. 7:03includes built-in AI credits, web
220. 7:05scraping already connected, and they
221. 7:08handle the updates, backups, and
222. 7:09security for you. You just go to the
223. 7:11OpenClaw landing page on Hostinger,
224. 7:13click get OpenClaw, choose your plan,
225. 7:16and by the way, if you do the 12- or
226. 7:1824-month option, it's even cheaper per
227. 7:20month. Then, during setup, you choose
228. 7:22your AI provider, pick your
229. 7:24communication channel like Telegram or
230. 7:26WhatsApp, and launch. And then, once
231. 7:28it's live, your agent runs 24/7. AI
232. 7:31agents like this used to feel very
233. 7:34gatekept. Like, only the really
234. 7:36technical people or big enterprises were
235. 7:38able to use them, but now it's pretty
236. 7:39cool how easy it's become to set it up
237. 7:41and use it for yourself at home. So, if
238. 7:44you've wanted your own AI assistant
239. 7:45without the headache of managing servers
240. 7:48manually, this is probably the easiest
241. 7:50way I've seen to do it. Check it out at
242. 7:52the link in the description box, and to
243. 7:53save even more money, use my code Matt
244. 7:56Wolfe for an additional 10% off. And
245. 7:58thank you so much to Hostinger for
246. 7:59supporting my channel and sponsoring
247. 8:01this portion of today's video. Now,
248. 8:03before I go any further on this, I do
249. 8:04want to give credit where credit is due.
250. 8:06This whole LLM knowledge base idea came
251. 8:08straight from Andrej Karpathy. I
252. 8:10specifically took the idea of using
253. 8:12Obsidian as the front end. Obsidian sort
254. 8:14of helps organize and easily read
255. 8:17markdown files. I'm just sort of
256. 8:19extrapolating off of this idea and
257. 8:21adding my journaling element and my CRM
258. 8:23element to the wiki concept that Andre
259. 8:26laid out here. Now, in order to build
260. 8:27this, you're going to need a couple
261. 8:28tools. I'm going to build this in Codex
262. 8:30here. This has been sort of my IDE of
263. 8:32choice lately to do coding and projects
264. 8:35like this. This is free to download, and
265. 8:37you get a certain amount of usage on the
266. 8:38free ChatGPT plan, but you're going to
267. 8:40get the most out of it if you do end up
268. 8:42upgrading. I'm also using Obsidian here.
269. 8:45Again, this is just a giant markdown
270. 8:47organizer and reader. This is totally
271. 8:49free to get. You can find it over at
272. 8:51obsidian.md.
273. 8:52You're also going to want the Obsidian
274. 8:54web clipper. So, if you're on the
275. 8:56obsidian.md website, I can scroll all
276. 8:58the way to the bottom here, and you can
277. 8:59see there's a link for web clipper.
278. 9:01Click in there, click add to Chrome, and
279. 9:03you'll get this little Obsidian web
280. 9:05clipper that you can see here in my
281. 9:07Chrome. And if I click this, this is
282. 9:08what automatically creates a new
283. 9:11markdown file, a new note for whatever
284. 9:13page you're on. And the cool thing that
285. 9:14I really like about this Obsidian web
286. 9:16clipper is that it automatically pulls
287. 9:18the transcripts from any YouTube video.
288. 9:20So, if I come over here to YouTube and
289. 9:22click on one of my recent YouTube
290. 9:24videos, and then I come up to the
291. 9:25Obsidian web clipper, it'll take a
292. 9:26second to load, but it will eventually
293. 9:28load the entire transcript for this
294. 9:31whole video straight into Obsidian, and
295. 9:33that makes it really easy to inject any
296. 9:35YouTube video or any article you find
297. 9:38around the web directly into your
298. 9:40Obsidian vault. So, once you have
299. 9:42Obsidian installed on your computer,
300. 9:44create a brand new vault, and I'm going
301. 9:46to call it second brain, and I'm going
302. 9:48to save it in a folder on my computer
303. 9:49called second brain. Now, I'll open
304. 9:52that, and we'll create this new vault,
305. 9:53and you can see I have a fresh blank
306. 9:56vault with nothing in it yet except for
307. 9:58a little welcome message. Now, it's
308. 10:00important to remember where on your
309. 10:02computer you just saved this vault, cuz
310. 10:04that's going to be necessary in the next
311. 10:05step here. I'm going to delete this
312. 10:07welcome message. It's not going to be
313. 10:08necessary, and now we have a purely
314. 10:10empty clean vault. So, for the next
315. 10:13step, I'm going to jump into Codex, and
316. 10:15we're going to actually build the dang
317. 10:16thing. So, over on the left here inside
318. 10:19of Codex, I'm going to click on add new
319. 10:21project, and then I'm going to select
320. 10:22use an existing folder. It's going to
321. 10:24open up my browser here to pick the
322. 10:26folder, and I'm going to go to the exact
323. 10:28folder that we just set our Obsidian
324. 10:31vault up with. So, for me, it's this
325. 10:32second brain folder that I created here,
326. 10:35and we'll go ahead and open that, and
327. 10:36then you can see I now have a project
328. 10:38over here called second brain. So, to
329. 10:40start this off, we're going to build the
330. 10:41basic bones of our wiki. And luckily,
331. 10:43Andre Karpathy generously gave us this
332. 10:46GitHub page that explains exactly how
333. 10:49the wiki architecture works. So, the
334. 10:52initial sort of hard part of building
335. 10:54the wiki is already figured out for us.
336. 10:56We can just take this URL to this GitHub
337. 10:58post here, open up Codex, make sure
338. 11:00we're in our second brain project folder
339. 11:03here, and giving it the prompt build out
340. 11:05architecture based on Karpathy's LLM
341. 11:07wiki here. I'm linking to that page on
342. 11:09GitHub that we were just looking at, and
343. 11:10then I said the current second brain
344. 11:12folder is the folder that Obsidian is
345. 11:14connected to. It is currently empty, so
346. 11:15we're building from scratch. And let's
347. 11:18go ahead and let it build out the sort
348. 11:20of architecture bones for us based on
349. 11:22what Karpathy's already figured out. All
350. 11:24right, so it worked for about 5 minutes.
351. 11:26It actually built out a whole bunch of
352. 11:29extra files that it didn't need to
353. 11:31build. I don't know why it created 51
354. 11:33files. The architecture is actually
355. 11:35supposed to be pretty small for this.
356. 11:37So, I literally prompted it, "Please
357. 11:38remove all the extra crap and just build
358. 11:40what's explicitly called for in
359. 11:42Carpathy's game plan." And it says,
360. 11:44"Done. I pruned it back to the minimal
361. 11:45Carpathy game plan." And now we just
362. 11:47have these files built in. If we pop
363. 11:50open Obsidian here, you could see we've
364. 11:52got just the folders we need. We have
365. 11:53the raw folder, we have the wiki folder,
366. 11:56we have our agents.md file, our index.md
367. 12:00file, and our log.md file. We can see
368. 12:03here exactly what each of these is for.
369. 12:05The raw folder is for the immutable
370. 12:07source material. This is where the
371. 12:08original stuff goes. Raw/assets, this is
372. 12:11for optional local Obsidian attachments.
373. 12:13You got the wiki. This is the
374. 12:15AI-generated markdown files that it's
375. 12:17pulling from the raw content that we're
376. 12:20inputting. You have the agents.md file,
377. 12:22which basically explains how this whole
378. 12:24thing works. So, we can see it's got the
379. 12:26operations. When the user adds a source
380. 12:28and asks LLM to process it, it does all
381. 12:30these things. When the user asks a
382. 12:32question, it queries it this way. So, it
383. 12:34basically tells it how this agent should
384. 12:36operate. You have the index.md file.
385. 12:39This is basically the catalog of
386. 12:41everything that's in the wiki. And then
387. 12:42you have the log file, where whenever
388. 12:44you make updates or changes or add
389. 12:45things, it updates the log file. Super,
390. 12:48super simple. We're starting bare-bones
391. 12:50here. If I look directly in the folder,
392. 12:52we just have what you see inside of
393. 12:54Obsidian. So, now I'm going to make sure
394. 12:56that my Obsidian web clipper is dialed
395. 12:58in. So, I'll go ahead and click on this.
396. 13:00We'll click on settings. Make sure you
397. 13:02add the name of your vault right here
398. 13:04under the vault list. If you're in
399. 13:05Obsidian down in the very bottom left
400. 13:07corner down here, you can see this is
401. 13:09the name of the vault. So, make sure
402. 13:11it's the same name exactly. And then
403. 13:13over under default, you've got the
404. 13:14templates over here. Click on the
405. 13:16default template and make sure that you
406. 13:18select that Second Brain vault or
407. 13:20whatever you titled it. And then I'm
408. 13:21having mine pull in these properties.
409. 13:23The source title, the source URL, the
410. 13:25date that it was created. That's the
411. 13:26date that I'm saving it in the web
412. 13:28clipper, not the date that the article
413. 13:29was actually written, and it's adding an
414. 13:31automatic web clip tag to it. And then
415. 13:34for the note content, it's just pulling
416. 13:35in the content. A lot of this might
417. 13:37actually be set for you by default, but
418. 13:39if it's not, this is what it should look
419. 13:40like. Under note location, we're going
420. 13:42to change this to just say raw, because
421. 13:45that is the folder inside of our
422. 13:47Obsidian vault that we want it to dump
423. 13:49it inside of. All right, so I can close
424. 13:51out of this, and for the very first
425. 13:52thing I'm going to ingest, might as well
426. 13:54ingest the instructions for how to build
427. 13:57one of these wikis. I know it's very
428. 13:58meta, but I want it inside of my wiki.
429. 14:01I'll click on my little Obsidian clipper
430. 14:03button, and you can see the source
431. 14:04title, LLM Wiki, we've got our source
432. 14:06URL, the date I'm pulling this in, and
433. 14:08the tags for web clip. And then here is
434. 14:10all of the content of this page here.
435. 14:12We'll go ahead and click add to
436. 14:13Obsidian, and we can see it added it
437. 14:15directly inside of the raw folder here
438. 14:18inside of Obsidian. Now, nothing's going
439. 14:20to happen automatically. We actually
440. 14:22need it to tell it to process the files
441. 14:24inside of raw for anything to actually
442. 14:26happen. But let's add a few more things.
443. 14:28I'm going to look through my YouTube
444. 14:29history and ingest some of the recent
445. 14:31videos that I've watched, like this
446. 14:33video called how to trick your brain
447. 14:35into becoming so disciplined your
448. 14:36friends will be shocked by your success.
449. 14:38I love a lot of psychology and mindset
450. 14:40type videos, so if I want to ingest
451. 14:41this, I come up to my web clipper, and
452. 14:43you can see it's going to pull in the
453. 14:45entire transcript here. So I'm going to
454. 14:47go ahead and click add to Obsidian, and
455. 14:49once again, we've got another file here
456. 14:51under raw. Now, there's one issue that's
457. 14:53going to pop up when I pull in YouTube
458. 14:55videos, is it's not going to properly
459. 14:57know the channel name, because it's not
460. 14:59automatically pulling it in into any of
461. 15:01these properties. But I can go to codex
462. 15:03here and give it some additional
463. 15:04instructions. When I save a video from
464. 15:07YouTube using the Obsidian web clipper,
465. 15:10and then you go and process the files,
466. 15:13make sure it also pulls the channel name
467. 15:16from YouTube and adds it as one of the
468. 15:18front matter fields. All right, so let's
469. 15:20go ahead and do that. So now let's go
470. 15:22ahead and do a quick test. We've got two
471. 15:23source files in here. So I'm going to
472. 15:25jump into Codex and go ahead and tell it
473. 15:27to process the files inside the raw
474. 15:30folder. Let's see how well it does right
475. 15:32now. All right, so it took about 3
476. 15:34minutes to process and it created a few
477. 15:37new sections. So let's just go ahead and
478. 15:39pull open Obsidian here and we can see
479. 15:41it left the original source material
480. 15:43here, but then it started to build out
481. 15:45the wiki of everything else. So we've
482. 15:48got our compounding knowledge base which
483. 15:50was clearly pulled from the explanation
484. 15:52from Andre, discipline without
485. 15:54willpower. This was pulled from this
486. 15:56channel, Aaron Miller study. Let me just
487. 15:59double check that it got the channel
488. 16:00name correct. Yep, Aaron Merrill study,
489. 16:03environment design. We can see this was
490. 16:05from the source discipline without
491. 16:07willpower, which was one of the concepts
492. 16:09that it saved, which came from this
493. 16:11original video that we saved. Identity
494. 16:13led goals, LLM wiki, temporal
495. 16:15discounting, and temptation bundling. So
496. 16:18we can see our wiki is starting to get
497. 16:19built out we have our index here, our
498. 16:21various sources, the LLM wiki and the
499. 16:24discipline without willpower. It
500. 16:25actually renamed it cuz it was
501. 16:27originally called how to trick your
502. 16:28brain into becoming so disciplined your
503. 16:30friends will be shocked by your success,
504. 16:31but it decided discipline without
505. 16:33willpower was a better name for it. We
506. 16:35can see the concepts here and it's
507. 16:37starting to build out. And if we look in
508. 16:38our log, we can see what it registered
509. 16:40in our log so far since we started
510. 16:42building this. One thing that I actually
511. 16:44like to do as this gets bigger and
512. 16:46bigger is you've got this graph view
513. 16:48here that starts really small when you
514. 16:50first build it and over time you'll see
515. 16:52this build out and build out and things
516. 16:54get more interconnected with each other
517. 16:56and it just gets really cool over time.
518. 16:58Now, I'm going to go seed this with some
519. 17:00more content. I'm going to go through my
520. 17:01watch history and pull in some of the
521. 17:03other videos that I watched recently,
522. 17:05using your money to be happier, the art
523. 17:06of tripod filmmaking, how to become
524. 17:08addicted to doing hard things, if you
525. 17:10think you're too busy watch this, how to
526. 17:12become a lucky person, and then build
527. 17:14your own self-improving AI wiki in 11
528. 17:16minutes. I know that's very meta, but
529. 17:18let's go ahead and import that. So, I'm
530. 17:20just going to go through and inject
531. 17:21every single one of these like we just
532. 17:23saw. I'm going to let it process them
533. 17:25all, and then they'll all be in the
534. 17:26Wiki, and then we'll move on to the next
535. 17:28steps, which are building out the
536. 17:30journal and the CRM elements that I
537. 17:32mentioned earlier. Okay, so it's done
538. 17:34injecting all of those videos that I
539. 17:36just saved. It took about 6 minutes
540. 17:38here, and this is what my Obsidian looks
541. 17:39like now. You can see all of the assets
542. 17:41of stuff that I ingested into it, and
543. 17:44the Wiki is getting built out quite a
544. 17:46bit more. We've got our index here, and
545. 17:48as you can see, the index is also
546. 17:50getting built out more as well. If I
547. 17:52click into like Hermes agent here, we
548. 17:55can see we've got key ideas from this
549. 17:57original video plus related content
550. 17:59inside of our Wiki. So, Codex
551. 18:01capabilities, I click on this one, and
552. 18:03it jumps to the video from Riley Brown
553. 18:05and the details around that one. Again,
554. 18:07this is the very, very simple, basic
555. 18:09setup of Carpathys' LLM Wiki. Now, if I
556. 18:13come into Codex again, we can
557. 18:15essentially chat with the Wiki. So, I
558. 18:17come to my second brain folder, click on
559. 18:19new chat, and I can ask questions. Like,
560. 18:21what are some tips for motivation when I
561. 18:23don't feel like doing the hard task
562. 18:25today? I know I saved a couple videos
563. 18:27about this exact topic, and we can see
564. 18:29it's already saying, "I'll treat this as
565. 18:30a Wiki query. First, I'm checking the
566. 18:32vault index, then I'll answer from
567. 18:34anything already captured and add the
568. 18:36reusable bit back into the Wiki if it
569. 18:38isn't there yet." Here's our final
570. 18:39response. When you don't feel like doing
571. 18:41the hard task, don't wait for motivation
572. 18:42to arrive first. Treat it as a task
573. 18:44design problem. Make the first few
574. 18:46minutes smaller, easier, and more
575. 18:47rewarding. Try this. Gives me a handful
576. 18:49of tips, and this was all pulled and
577. 18:51grounded from the Wiki, but it also
578. 18:54updated the Wiki based on the question
579. 18:56that I asked. You can see that it
580. 18:57changed the index.md file, the log.md
581. 19:00file, and the Wiki motivation for hard
582. 19:03tasks. So, opening up my Obsidian vault
583. 19:05here again, looking in my log, we can
584. 19:07see that it actually logged this query,
585. 19:09motivation when avoiding a hard task,
586. 19:11answer to query about motivation, and it
587. 19:13even updated the index with it. And it
588. 19:15created motivation for hard tasks and
589. 19:17linked back to the original sources that
590. 19:19it found this information from. So, as
591. 19:22you ask questions, the wiki further and
592. 19:24further and further builds out based on
593. 19:26the questions you were asking. Now,
594. 19:28there's a few things that I want to do
595. 19:29to clean this up a little bit and make
596. 19:31it slightly more useful for me cuz right
597. 19:33now, once it processes something, it
598. 19:35just leaves it in this raw folder, and
599. 19:37this is just going to build up and build
600. 19:38up. And so, what I want to do is under
601. 19:40this raw folder here, I'm going to go
602. 19:42ahead and create a new folder, and I'm
603. 19:44going to call it processed. Whenever it
604. 19:46processes one of these files and adds it
605. 19:48to the wiki, I want it to move it to the
606. 19:50processed folder, so I know that that
607. 19:52has already been ingested. So, now that
608. 19:54I've got this processed folder, I can
609. 19:56simply come down to my agents file here
610. 19:58and then tweak what happens when the
611. 20:00file is processed. So, if I come down
612. 20:02here, we've got operations ingest when
613. 20:04the user adds a source and asks the LLM
614. 20:06to process it, read the source from raw,
615. 20:08create or update wiki pages, update
616. 20:10relevant entity concept topic overview
617. 20:12synthesis or comparison pages, update
618. 20:14index.md, append an entry to log.md.
619. 20:18Well, now I can just add a number six
620. 20:20and say move the source file from the
621. 20:22root raw directory to raw/processed.
622. 20:27By adding that extra bit to the little
623. 20:29prompt here, now it's going to go
624. 20:30through all these steps, but then move
625. 20:32it into the processed folder. It also
626. 20:34misunderstood me when I said to add the
627. 20:36channel name. It thought I wanted it to
628. 20:37add the channel name to the actual wiki
629. 20:40generated page, but I wanted it to add
630. 20:42the channel name to the original source.
631. 20:44That's what makes the most sense to me.
632. 20:46So, I'm going to come to my agents
633. 20:47section here and just tweak that as
634. 20:50well. So, right now it says for YouTube
635. 20:51videos clipped with Obsidian Web
636. 20:53Clipper, also open or inspect the
637. 20:54YouTube source URL and add the channel
638. 20:56name to the generated wiki page front
639. 20:58matter. But instead of that, I'm going
640. 20:59to say add the channel name to the
641. 21:02original source page front matter. I I
642. 21:05want it to link back to the original
643. 21:06source. So, I'm going to add a step
644. 21:08right after step three here and say
645. 21:10cross-link any wiki pages generated or
646. 21:13updated to the original source page.
647. 21:15Basically, I don't want these pages
648. 21:16orphaned. If there's a new wiki page
649. 21:18here, I want it to link to the original
650. 21:21page here. So, that's the manual way to
651. 21:23update the agents.md file, but you can
652. 21:25also do it by prompting it inside of
653. 21:28Codex. So, if I come to my second brain
654. 21:30project, create a new chat here, I can
655. 21:32give it instructions on additional
656. 21:34things that I want to happen. So, I
657. 21:35mentioned my journal and I mentioned my
658. 21:37CRM. So, let me go ahead and build the
659. 21:40bones for that here. I will close these
660. 21:42folders to clean everything up. I will
661. 21:43create a new folder called journal and a
662. 21:46new folder called CRM. Now, I can come
663. 21:49to Codex and say update the agents.md
664. 21:52file to handle these items. Number one,
665. 21:55if I start a chat with journal, add the
666. 21:57text of that chat and subsequent
667. 21:59conversation as a new md file within the
668. 22:02journal folder. The entire conversation
669. 22:03should be added to the markdown file.
670. 22:05Create an index file in the journal
671. 22:07folder that's similar to the wiki index
672. 22:09file. Each new journal entry gets added
673. 22:11to the index file. Decide on a short
674. 22:13title for the journal entry based on the
675. 22:15contents of the journal and use the date
676. 22:17and the title as the journal entry file
677. 22:19name. Add the date and title to the
678. 22:21index and link to the entry. Also, log
679. 22:24the journal entry title and short
680. 22:25summary in the log.md file. Your
681. 22:27response to my journal entry should be
682. 22:29grounded in content from the wiki in the
683. 22:31same way you view the index and respond
684. 22:33to my chat questions based on what's in
685. 22:34the wiki. Provide advice and insights to
686. 22:36my journal entries based on what's
687. 22:38available in the wiki, as well as your
688. 22:40own LLM knowledge. Provide helpful
689. 22:42advice, insights, guidance, tactics, and
690. 22:43ideas using what you know, along with
691. 22:45what's available from the wiki, past
692. 22:47journal entries, and the CRM. So, when I
693. 22:49journal, I want it to look in the wiki,
694. 22:51find information that's helpful to what
695. 22:52I just journaled on. I want it to look
696. 22:54in past journal entries to see if
697. 22:56there's anything relevant I've journaled
698. 22:57on in the past, and I want it to look at
699. 22:59my CRM and see if there's conversations
700. 23:01I've had with people about what I'm
701. 23:03journaling about, too, for the CRM. If I
702. 23:05tell you I'm giving you information for
703. 23:07the CRM, either update the person in the
704. 23:09CRM or add the person to the CRM. CRM
705. 23:12file should always be a person's name. I
706. 23:14will share details about a person, their
707. 23:15name, contact details I have for them,
708. 23:17details about where or how we met,
709. 23:19things that I know about them, etc.
710. 23:21Creator, update the contact record in
711. 23:22the CRM with whatever details I give
712. 23:24you. In the CRM folder, create an index
713. 23:26file similar to our other index files
714. 23:28with the name of the people in the CRM
715. 23:30listed in alphabetical order and a short
716. 23:31bio of what information I have about
717. 23:33that person. This will allow me to ask
718. 23:35questions about contacts that are inside
719. 23:38the CRM. I want it to update these two
720. 23:39things in our agents.md file, which will
721. 23:42make it so that whenever I chat with my
722. 23:44second brain project here in Codex, it's
723. 23:46either one going to answer the question
724. 23:49that I asked it using the sort of query
725. 23:52task that's already built into the
726. 23:53agent, it's going to handle it as a
727. 23:55journal if I preempt it with journal
728. 23:58here, or if I tell the chatbot that it's
729. 23:59for the CRM, it will update the CRM
730. 24:02section. So, I'm going to let it go
731. 24:03ahead and update our agents.md file,
732. 24:06create the various index files, and that
733. 24:08should build out the system for these
734. 24:10elements. Okay, so we can see that it
735. 24:12updated the agents.md file with our
736. 24:14journal rules, our CRM rules, and if we
737. 24:17open up Obsidian once again, I can open
738. 24:19my journal folder and you can see we've
739. 24:21got an index with date entry and
740. 24:23summary, and if I open up my CRM, we've
741. 24:25got our index with the person and the
742. 24:27summary, and where most of the updates
743. 24:29happened was this agents.md file. So,
744. 24:32now below the wiki section, we have a
745. 24:34journal section. It built all the rules
746. 24:36that I just asked it to build for me,
747. 24:38and then we also have our CRM section.
748. 24:40So, it created all the rules for the CRM
749. 24:42section now. Let's jump into Codex one
750. 24:45more time and let's test this out. I'm
751. 24:47just going to say add to CRM, and then
752. 24:50I'll put Matthew Berman, and then I'll
753. 24:53just put met at Qualcomm event in 2024,
754. 24:58hung out at CES in 2025,
755. 25:02got lunch during TechCrunch disrupt in
756. 25:052025, and then I would obviously add
757. 25:08whatever sort of contact details I have
758. 25:10about him, but no point in doing that on
759. 25:12camera. I'll just tell it to add that to
760. 25:14the CRM, and we can see updating the CRM
761. 25:16record for Matthew Berman, creating a
762. 25:17new record, and it says, "Great, I
763. 25:19created that." If I look in my Obsidian
764. 25:21here, we can see under CRM, we now have
765. 25:24a Matthew Berman in CRM here with our
766. 25:26summary, no contact details provided,
767. 25:28how we met, relationship context, and
768. 25:31you know, other stuff that's still empty
769. 25:33right now. And then if I look at my
770. 25:34index, we can see it added the person,
771. 25:36Matthew Berman, and a quick summary
772. 25:38there. It also added it to the log down
773. 25:40here that the CRM was updated. But the
774. 25:42idea being is, let's say I'm going to a
775. 25:45conference, and I bump into somebody,
776. 25:47and I'm like, "Oh, shoot, where did we
777. 25:48meet?" I can jump into my second brain
778. 25:49and say, "Where did I meet Matthew
779. 25:51Berman again?" I'll check the vault and
780. 25:53CRM records. You met Matthew Berman at
781. 25:55Qualcomm event in 2024 according to your
782. 25:57CRM notes. So, a really good way to
783. 25:59reference information later. I'm going
784. 26:01to go ahead and make a pretend journal.
785. 26:03So, what I'll do is I'll click on a new
786. 26:05chat inside of my second brain project
787. 26:07here, and I'll start it with journal,
788. 26:09and then I'll go to another line, and
789. 26:11I'll just brain dump some thoughts. So,
790. 26:13this is just a simple demo journal
791. 26:16entry, but you know, it's got some
792. 26:18reality to it. I have a ton of ideas for
793. 26:19videos, but I decide not to make them
794. 26:21because I'm worried they won't get a
795. 26:22good view count. I also talk about how I
796. 26:24know how to write titles that will get a
797. 26:26lot of clicks, but I also hate the whole
798. 26:28clickbait thing, so I end up making
799. 26:30titles that I know will get less clicks
800. 26:33because they're more literal to what the
801. 26:35video is about, and I constantly grapple
802. 26:37with making videos that will get a lot
803. 26:39of clicks, but knowing I'll get hate in
804. 26:41the comments because they're too
805. 26:43click-baity versus making videos that
806. 26:46the title is very literal, so I won't
807. 26:48get the hate comments, but I'll also get
808. 26:50a lot of less views. And that's
809. 26:51something that I grapple with, and I
810. 26:53know pretty much every YouTuber grapples
811. 26:55with. And so, that's my short journal
812. 26:57entry. So, let's go ahead and submit
813. 26:58that and see what happens. I'll treat
814. 27:00this as a journal entry and grab the
815. 27:02response in the vault first. Going to
816. 27:04read the indexes, look for relevant
817. 27:05prior notes around creating video
818. 27:07attention, then I'll write the journal
819. 27:08file before answering. The vault has
820. 27:10several creator strategy pages. I'm
821. 27:12pulling those in rather than answering
822. 27:13from a blank slate. There aren't prior
823. 27:15journal entries yet, and the CRM doesn't
824. 27:17appear relevant to this one. So, it's
825. 27:18just pulling from the wiki. Okay, and
826. 27:20check out this journal response here. I
827. 27:22think there are two fears braided
828. 27:23together here, creative integrity and
829. 27:25channel safety. One part of you does not
830. 27:27want to betray the video with a title
831. 27:28that feels gross. Another part is afraid
832. 27:30that a few underperformers will teach
833. 27:32YouTube not to trust you. The creator
834. 27:33notes in the vault push back on the
835. 27:34second fear. YouTube value of death and
836. 27:36creator persistence both frame visible
837. 27:38metrics as lagging indicators. A few
838. 27:41lower view videos are not a downward
839. 27:42spiral, blah blah blah. So, it is
840. 27:44actually grounding the response to my
841. 27:46journal in what is existing in the wiki.
842. 27:49And then it actually goes on to just
843. 27:50give a response that ChatGPT would
844. 27:52normally give, but it's pulling from
845. 27:54both the knowledge built into the LLM as
846. 27:57well as the knowledge in the wiki. So,
847. 27:59that's all working exactly as intended.
848. 28:02And then if I want to pop open Obsidian
849. 28:04here, I can open my journal folder, and
850. 28:06we can see here is my journal entry.
851. 28:08Here is what the assistant replied with.
852. 28:10Here's some synthesis, and here's some
853. 28:12related content about what I'm
854. 28:14journaling about. If I look at my index
855. 28:16here, it automatically logged the date,
856. 28:18the entry title, and a quick like one
857. 28:20paragraph summary of what my journal
858. 28:22entry is about. Now, I want some of this
859. 28:24to happen on autopilot. I don't want to
860. 28:26have to jump into Codex constantly and
861. 28:29tell it to process everything that I
862. 28:31saved in here. And there's an easy
863. 28:32solution for that. Before I show you,
864. 28:34I'm going to do one thing. I'm going to
865. 28:36go ahead and clean up my existing
866. 28:38Obsidian wiki because, remember, I want
867. 28:40these to all get moved into processed,
868. 28:42and I also want it to save the YouTube
869. 28:45channel name up here in the front
870. 28:46matter. So, I'm going to tell it,
871. 28:47"Please reprocess all of the files in
872. 28:51the raw directory following the recent
873. 28:55updated instructions on how to process
874. 28:59them." So, I'm going to let that
875. 29:00process. It's going to clean up my wiki
876. 29:01real quick. All right, now that that's
877. 29:03finished and my Obsidian vault is
878. 29:05cleaned up and we can see all of my
879. 29:06videos have been moved to process and it
880. 29:08added the channel name to the original
881. 29:10videos here. Let's automate some of
882. 29:12this. So, in Codex here, you've got a
883. 29:14feature called automations. This is
884. 29:16where you can set it up to do recurring
885. 29:17tasks. So, if I click into automations
886. 29:20here, select new automation. For
887. 29:21automation title, we'll call it process
888. 29:24second brain raw files. For work tree,
889. 29:27I'm going to set it on local so it runs
890. 29:28directly in the selected project. For
891. 29:30our project, we're going to select
892. 29:31second brain. Here, we'll select when we
893. 29:33want it to run. I'm going to go ahead
894. 29:34and set mine to hourly, but you can do
895. 29:36it at whatever cadence you want. And I'm
896. 29:38just going to say if there are any
897. 29:41unprocessed files inside the raw
898. 29:43directory, please process them now. For
899. 29:46the model, I'm going to set it as a
900. 29:47GPT-5.5.
901. 29:49I recommend just using the strongest
902. 29:51model you have available. I'm going to
903. 29:52set it on high reasoning and I will
904. 29:55create this automation. Now, it's going
905. 29:56to run every hour, see if there's
906. 29:58anything in my raw folder that's
907. 30:00unprocessed, and then it will process
908. 30:02it. And that's it. That's the whole
909. 30:04process now. Whenever I come across
910. 30:05stuff I want to save, I just use the
911. 30:07Obsidian web clipper and clip it into my
912. 30:09raw folder automatically. And every
913. 30:12hour, it's going to ingest that and turn
914. 30:13it into one of the wiki pages. If I want
915. 30:15to add somebody to my CRM, I just open
916. 30:17up Codex, create a new chat inside of
917. 30:19this project, and add the CRM details.
918. 30:21If I feel like journaling right now, I
919. 30:23can journal straight into my second
920. 30:24brain, and it will ground the response
921. 30:27in what's available in my wiki, past
922. 30:29journals, and within my CRM. If you want
923. 30:30an extra layer of backup, you can also
924. 30:33go to GitHub, create a new repository on
925. 30:35GitHub. I'm going to go ahead and call
926. 30:37this one second brain. I'm going to set
927. 30:39this to private, so it's only available
928. 30:41to me, and I'll create the repository.
929. 30:43Now, if I copy this URL, jump back over
930. 30:45to Codex here, I can say commit this
931. 30:48current version to my private GitHub
932. 30:52repo here. Pasting the URL of the
933. 30:54private GitHub repo I just created, and
934. 30:56then go ahead and submit it. Now, I've
935. 30:58previously attached the GitHub plugin,
936. 31:00so it should just work out of the box.
937. 31:02You can see I'm already synced with
938. 31:04GitHub, but if you haven't done that
939. 31:05already, just add the GitHub plugin and
940. 31:07go through the motions to get that set
941. 31:09up, and you should be good to go. And it
942. 31:11went ahead and pushed it to GitHub, so
943. 31:12if I open my browser here and refresh,
944. 31:15we can see everything I've created is
945. 31:17all saved on GitHub now. If I jump back
946. 31:19into Codex, I can go to my automation
947. 31:21and edit this automation and say if
948. 31:24there are any unprocessed files in the
949. 31:25raw directory, please process them now.
950. 31:28Once everything is processed, please
951. 31:29commit and push the current version of
952. 31:30the directory to the main branch on
953. 31:32GitHub. So, now it's going to process
954. 31:33everything in the raw directory, and
955. 31:35then once it's done processing, it's
956. 31:36going to update GitHub so that backup is
957. 31:39constantly happening every hour. And
958. 31:41there you have it. There's the whole
959. 31:42second brain process. Not only do you
960. 31:44have a wiki of all of the information
961. 31:46you're finding and saving from around
962. 31:47the internet, but now you have a journal
963. 31:49and a CRM that's built on top of it as
964. 31:51well. And if you ever want to tweak how
965. 31:53it operates, you just open up Obsidian,
966. 31:55which is your visibility layer to see
967. 31:57how everything is built, and you go into
968. 31:59agents.md, and you just tweak the
969. 32:02instructions. This is all just prompts
970. 32:04at the end of the day. You just change
971. 32:05how it gets prompted. And in the short
972. 32:07amount of time that we've been working
973. 32:08on it, this is what our graph view looks
974. 32:10like now, and we can start to see all of
975. 32:12these things interconnect with each
976. 32:13other a little bit more. Do this for a
977. 32:16few days and a few weeks, and the next
978. 32:17thing you know, you have a vault that
979. 32:19looks like this. Uh yeah, pretty insane,
980. 32:22crazy vault that just has a ton of
981. 32:24information saved inside of it. So, I
982. 32:26know that video was long. There was a
983. 32:28lot of details. I wanted to make sure it
984. 32:29was very clear and you got the whole
985. 32:31process, and I didn't skip any steps,
986. 32:33but I want to show you how I've been
987. 32:35building this sort of second brain
988. 32:36concept that I can journal on top of
989. 32:38that I have a CRM on top of where all of
990. 32:40the wiki elements are interconnected.
991. 32:42You can even dial it in more by building
992. 32:45in separate folders and telling it to
993. 32:47break out people from different pieces
994. 32:49of content and break out companies from
995. 32:51different pieces of content and really,
996. 32:53really dial in that wiki more and more
997. 32:55and more. But really, really cool
998. 32:57concept. All you really need is Obsidian
999. 32:59and Codex. Anthropic's co-work or Claude
1000. 33:02code also works. Whatever your sort of
1001. 33:05front-end chat platform of choice is.
1002. 33:07I've been really liking Codex lately, so
1003. 33:10that's what I've been using. But yeah,
1004. 33:11go build it. It's really, really cool
1005. 33:13and a lot simpler than you think and
1006. 33:15over time it just gets smarter and
1007. 33:17smarter and smarter and more and more
1008. 33:18powerful. So, that's what I got for you.
1009. 33:20This isn't like my normal videos. I
1010. 33:22normally make end of week news
1011. 33:24breakdowns where every single Friday I
1012. 33:26break down all of the news that happened
1013. 33:28in the AI world for the week. I drink
1014. 33:29from the fire hose all week, keep up
1015. 33:31with all of the news, keep myself
1016. 33:33completely looped in so that other
1017. 33:35people don't have to feel overwhelmed.
1018. 33:37I'll take on the overwhelm and just
1019. 33:39report what I think is the most
1020. 33:40interesting. I put those videos out
1021. 33:42every single Friday. If stuff like that,
1022. 33:44as well as tutorials like this, are
1023. 33:46something that interest you, maybe
1024. 33:47consider liking this video and
1025. 33:48subscribing to this channel. It really,
1026. 33:50really helps me out a lot. But again,
1027. 33:51that's what I got. Thanks for hanging
1028. 33:52out with me, nerding out with me.
1029. 33:54Hopefully, I'll see you in the next one.
1030. 33:55Bye-bye.