YouTube2Text

Graphify + Obsidian is INSANE: Build an AI Second Brain That Never Forgets — Transcript

by Cloud Codes · 1,824 words · 265 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This is an AI second brain that never
  2. 0:02forgets. Point it at a folder and your
  3. 0:04whole project turns into a graph your
  4. 0:06assistant can query because right now
  5. 0:08your coding assistant has amnesia. Every
  6. 0:10new chat it has read nothing. Ask it how
  7. 0:13your login flow works and it opens every
  8. 0:15file from scratch to figure it out. Do
  9. 0:18that across 200 files on every single
  10. 0:20question and you pay a token tax each
  11. 0:22time. And stuffing your entire repo into
  12. 0:24a giant context window just makes the
  13. 0:26answers worse. On April 3rd, Andre
  14. 0:29Carpathy posted a fix and the dev world
  15. 0:31copied it in a single weekend. Stop
  16. 0:34making the model rebuild your knowledge
  17. 0:35every query. Give it a wiki it keeps up
  18. 0:38to date itself. 2 days later, a
  19. 0:40developer shipped exactly that as a tool
  20. 0:42called Graphify. From nothing, it blew
  21. 0:44past 90,000 GitHub stars in about 3
  22. 0:47months. Pair it with Obsidian, and that
  23. 0:49graph becomes plain Markdown you own.
  24. 0:51So, here's the full picture. How
  25. 0:53graphify builds the graph, why Obsidian
  26. 0:56is its home, and the four ways it falls
  27. 0:58apart if you get lazy. And this was not
  28. 1:00some quiet launch. It turned into a
  29. 1:02genuine wave. That claude code plus
  30. 1:05Obsidian idea pulled over 50,000 likes
  31. 1:07on X in a single week. One post
  32. 1:09declaring notetaking dead hit 54,000 on
  33. 1:12its own. Carpathy's little gist alone
  34. 1:15picked up 14,000 stars and,900 forks.
  35. 1:18First, the actual problem Graphify is
  36. 1:20solving. The usual fix for a big code
  37. 1:23base is what's called rag. Chop your
  38. 1:25files into chunks, turn them into
  39. 1:26vectors, and search for the closest
  40. 1:28matches when you ask a question. That
  41. 1:30works great for pros. For code, it's
  42. 1:32shaky because the link between a
  43. 1:34function and the things that call it
  44. 1:36isn't about similar words. It lives in
  45. 1:38the call graph. You end up paying to
  46. 1:40load the whole library just to find one
  47. 1:42paragraph. Graphify flips that around.
  48. 1:44Instead of searching text on every
  49. 1:46question, it does the expensive reading
  50. 1:48once up front and squeezes your project
  51. 1:50into an explicit graph of entities and
  52. 1:53the relationships between them. After
  53. 1:55that, answering a question means walking
  54. 1:57the graph instead of rereading the
  55. 1:58files. It's exactly how a senior
  56. 2:00engineer works. Build the mental map
  57. 2:02once, then just follow it. And it isn't
  58. 2:05only code. Graphify pulls in your SQL
  59. 2:07schemas, your shell scripts, your R
  60. 2:10notebooks, your architecture PDFs, even
  61. 2:12recorded videos, and drops them into one
  62. 2:15graph. Application code, database
  63. 2:17schema, and infrastructure finally
  64. 2:19living on the same map instead of in
  65. 2:21five different heads. Actually using it
  66. 2:23is almost boring in the best way. Inside
  67. 2:26cloud code or codeex or cursor, you type
  68. 2:29/graphify and point it at a directory.
  69. 2:32It grinds through the folder and drops a
  70. 2:34graphify out folder right next to your
  71. 2:36code with the entire graph saved to
  72. 2:38disk. It supports every big assistant,
  73. 2:40not just one. Here's the part one
  74. 2:42genuinely respect. Graphify does not
  75. 2:45treat every file the same way. It runs
  76. 2:47three separate passes over your stuff
  77. 2:49and each one has a completely different
  78. 2:51story about where your data ends up.
  79. 2:53Pass one is your source code. Graphy
  80. 2:55runs it through tree sitter, the same
  81. 2:57kind of parser a compiler uses across 40
  82. 2:59languages. It's deterministic. There's
  83. 3:02no AI model involved and there's no
  84. 3:04network call. Your code stays on your
  85. 3:06machine and every link it finds gets
  86. 3:08tagged as a hard fact. Pass two is for
  87. 3:11recordings. Point it at a folder with
  88. 3:13audio or video, say a design meeting you
  89. 3:15screen recorded, and it transcribes them
  90. 3:17locally with faster whisper. An hour of
  91. 3:20talking becomes text right on your
  92. 3:21laptop and folds into the graph. Nothing
  93. 3:24gets uploaded. Pass 3 is the honest one.
  94. 3:27Documents and images, a PDF spec, or a
  95. 3:30photo of a whiteboard can't be read by a
  96. 3:32grammar. So, those get sent to your AI
  97. 3:34provider to be understood. That's the
  98. 3:36only pass that leaves your machine, and
  99. 3:38it goes to your own API key, not some
  100. 3:40middleman. Want it fully private? Run
  101. 3:43code only mode and skip this pass. And
  102. 3:45it refuses to sound sure about things it
  103. 3:47isn't. Every connection in the graph
  104. 3:49carries a label. Extracted means it's
  105. 3:51right there in your code, so trust it.
  106. 3:53Inferred means the model reasoned it
  107. 3:55from context. So probably ambiguous
  108. 3:58means the model itself raised a hand. So
  109. 4:00check before you build on it. Once the
  110. 4:02graph exists, it organizes itself.
  111. 4:05Graphify runs community detection and
  112. 4:07your modules just fall out of the shape
  113. 4:09of the code. O in one cluster, billing
  114. 4:11in another, infrastructure off on its
  115. 4:14own with the bridge files between them
  116. 4:15lit up. You didn't draw that map. The
  117. 4:18relationships did. Now the payoff at
  118. 4:20query time. You ask, "How does a user
  119. 4:23log in?" Instead of loading 50 files,
  120. 4:25Graphify walks the graph and hands back
  121. 4:28a short path. Three hops, zero files
  122. 4:30opened. Your assistant reads a few node
  123. 4:32summaries instead of half the repo. And
  124. 4:34querying isn't just one trick. Graphify
  125. 4:37explain hands you a plain English tour
  126. 4:39of any entity. Graphy path finds the
  127. 4:42shortest chain between two things, say
  128. 4:43auster and the database pool it secretly
  129. 4:46leans on. and deep mode goes hunting for
  130. 4:48the fuzzier inferred links a quick pass
  131. 4:50would skip right over. Graphy puts a
  132. 4:53number on all this about 71 times fewer
  133. 4:55tokens per query on a mixed codebase.
  134. 4:58Now that's their own benchmark on their
  135. 5:00own data and nobody outside has
  136. 5:01reproduced it yet. So take the exact
  137. 5:03figure with a grain of salt, but the
  138. 5:05direction is obviously right. Walking a
  139. 5:08map is cheaper than rereading the whole
  140. 5:09library every time. And that one graph
  141. 5:12isn't trapped inside the tool. Graphy
  142. 5:14exports it every way you'd want. a
  143. 5:16clickable HTML graph rendered with viz.s
  144. 5:19s plain JSON for scripts, a markdown
  145. 5:22report of your busiest files, a Neo4j
  146. 5:24database, even an MCP server, so any
  147. 5:27agent can call your codebase graph like
  148. 5:29a built-in tool. It also stays current
  149. 5:32without you babysitting it. Every file
  150. 5:34gets a chat 256 hash, so the second run
  151. 5:37only rereads what actually changed. Add
  152. 5:39a git hook and the graph rebuilds itself
  153. 5:42on every commit. Your assistant sees a
  154. 5:44map that matches the exact branch you're
  155. 5:46standing on. So, where does Obsidian
  156. 5:48fit? Obsidian is a notes app built on
  157. 5:51plain markdown files that live on your
  158. 5:53disc, not on someone's server. Around 1
  159. 5:56and a.5 million people use it, and its
  160. 5:58plugins just crossed 100 million
  161. 5:59downloads. It's the natural home for a
  162. 6:01brain you actually want to own. Point
  163. 6:03Graphifies Obsidian export at a vault,
  164. 6:06and your graph turns into real notes.
  165. 6:08Every entity becomes a page. Every
  166. 6:10relationship becomes a Wiklink. And
  167. 6:12Obsidian's graph view shows the whole
  168. 6:14thing as a web. You can wander through
  169. 6:16your codebase as a mind map. You can
  170. 6:18open on a Sunday. Then you close the
  171. 6:20loop with a plugin like Claudian, which
  172. 6:22drops Claude code straight into the
  173. 6:23Obsidian sidebar. Your vault becomes the
  174. 6:25agents working directory. It can read
  175. 6:27your notes, write new ones, and run
  176. 6:29commands without leaving the app your
  177. 6:31knowledge already lives in. And Graphify
  178. 6:33isn't the only door in. You don't have
  179. 6:36to hand over full write access on day
  180. 6:38one. Smart connections just surfaces
  181. 6:40related notes using embeddings. Cudge is
  182. 6:43the self-hosted run it all locally
  183. 6:44option. Copilot for Obsidian gives you
  184. 6:47vault aare chat and is already past
  185. 6:49100,000 users. Start light then give it
  186. 6:52more rope. Underneath all of this sits
  187. 6:54Carpathy's actual design and it's
  188. 6:56refreshingly simple. Three layers. A raw
  189. 6:59folder of sources the AI leaves
  190. 7:01untouched. A wiki layer the AI does own
  191. 7:04where it writes pages and keeps the
  192. 7:05cross links honest. And a schema file
  193. 7:08that tells the model how your wiki is
  194. 7:09laid out. Three operations keep that
  195. 7:11wiki alive. Ingest pulls in new sources
  196. 7:14and updates every page they touch. Query
  197. 7:17searches the wiki and answers you with
  198. 7:18citations. And lint runs a health check
  199. 7:20for contradictions, dead pages, and
  200. 7:23stale claims. Remember that word lint.
  201. 7:25It comes back to bite people. And what
  202. 7:27gets me is how old this idea actually
  203. 7:30is. Carpathy traces it back to 1945 to
  204. 7:33Vanavar Bush's memes. The dream of link
  205. 7:35trails between your documents. The links
  206. 7:37mattered as much as the pages
  207. 7:39themselves. We just lacked a tool
  208. 7:40patient enough to maintain them. Now the
  209. 7:43model does that upkeep for free. Okay,
  210. 7:45I've talked this up for 5 minutes, so
  211. 7:47let me be the annoying person in the
  212. 7:49comments for a second. The viral demos
  213. 7:51leave out the part where this whole
  214. 7:52thing rots if you ignore it. There are
  215. 7:54four cracks worth knowing about. Crack
  216. 7:56one is size. A vault with 50 tidy notes
  217. 8:00feels like actual magic. A vault with
  218. 8:022,000 notes turns into a search problem
  219. 8:04all over again because the model spends
  220. 8:06its context just navigating your
  221. 8:08structure instead of answering your
  222. 8:10question. The graph softens that. It
  223. 8:12doesn't erase it. Crack 2 is confident
  224. 8:15fiction. When the assistant writes up
  225. 8:16your notes, it cheerfully fills the gaps
  226. 8:19with things you never said in your own
  227. 8:20voice. The fix is a blunt line in your
  228. 8:23instructions. Do not add anything I
  229. 8:25didn't write. Without that rule, your
  230. 8:27second brain starts inventing memories.
  231. 8:29Crack three is feedback loops. The AI
  232. 8:32writes a page, then weeks later cites
  233. 8:34its own page as a source, then builds on
  234. 8:36top of that. Let it run unattended and
  235. 8:38you get AI quoting AI where small
  236. 8:40mistakes slowly harden into facts. Your
  237. 8:43brain fills up with its own echo. Crack
  238. 8:454 is the big one, and it's why that word
  239. 8:47lint mattered. Carpathy built a cleanup
  240. 8:50pass into the design on purpose, and
  241. 8:52almost every tutorial skips it. 5
  242. 8:54minutes a week, read what the agent
  243. 8:56wrote, prune the stale pages, kill the
  244. 8:58contradictions. Skip that and the whole
  245. 9:01thing decays into noise. Do it and it
  246. 9:03compounds into something real. So, my
  247. 9:06honest take, a second brain that never
  248. 9:08forgets is genuinely here and it's the
  249. 9:10best version we've had. But that memory
  250. 9:12cuts both ways. It also keeps every bad
  251. 9:15note and lazy guess unless you clean up
  252. 9:17after it. Own your markdown, keep it on
  253. 9:19your disc, and treat that weekly lint
  254. 9:21pass as the rent. And if you do one
  255. 9:23thing this week, start almost
  256. 9:25embarrassingly small. One folder, one
  257. 9:27graph, one vault. Let it earn your trust
  258. 9:30on a project you already understand
  259. 9:32before you feed it your entire life. The
  260. 9:34people who win with this aren't the ones
  261. 9:35with the sickest demo. They're the ones
  262. 9:37whose vault still tells the truth 3
  263. 9:39months from now. That's the whole build.
  264. 9:42A brain that finally keeps up with you.
  265. 9:44See you in the next one.

About this transcript

This page contains the full transcript of Graphify + Obsidian is INSANE: Build an AI Second Brain That Never Forgets by Cloud Codes, generated from the public captions YouTube serves with the video. The transcript has 1,824 words across 265 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.