YouTube2Text

How to Use Claude Code for FREE (2026) — Transcript

by Nick Saraev · 6,843 words · 983 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Hey, this is a video on how to use
  2. 0:01Claude Code completely free. So, Claude
  3. 0:02Code is the number one coding agent in
  4. 0:03the world right now, but it also costs
  5. 0:05between 20 to 200 bucks a month. It has
  6. 0:07pretty restrictive rate limits, as we
  7. 0:08know, and some people complain about
  8. 0:10quality fluctuations. In this video, I'm
  9. 0:11going to show you a free, fully
  10. 0:13open-source alternative method called
  11. 0:14free Claude Code that literally lets you
  12. 0:16use the exact same Claude Code
  13. 0:17interface, so the same terminal and the
  14. 0:19same commands, but with significantly
  15. 0:21more affordable back-end models that
  16. 0:23allow you to achieve somewhere between
  17. 0:2480 to 90% of the quality of Opus 4.7,
  18. 0:26but for between 2 to 5% of the cost. So,
  19. 0:29very high leverage. Let me show you how
  20. 0:30to do it. So, this is an app called
  21. 0:31Habitual that I just put together, and
  22. 0:33uh it's just a simple habit tracker app.
  23. 0:35What's funny is this would have cost me
  24. 0:37something like 5 to 10 dollars in in
  25. 0:38Anthropic credits. With Deep Seek Flash
  26. 0:41V4, which is the model that I was using
  27. 0:42inside of this free Claude Code uh
  28. 0:44configuration, I was able to build this
  29. 0:46for like 3 cents. So, that's not
  30. 0:48hyperbole. I literally did so for like
  31. 0:50several hundred times less money than I
  32. 0:52would pay to Anthropic. And it's not to
  33. 0:54say that this is the exact same quality
  34. 0:55of app that I would get if I built, you
  35. 0:57know, using Claude, but uh you know, the
  36. 0:59the coding models that are available now
  37. 1:01on platforms like OpenRouter, Nvidia
  38. 1:03NIM, and then Ollama, all three sources,
  39. 1:05which I'm going to run you guys through
  40. 1:06how to set up on your end, are actually
  41. 1:08quite powerful, and you can do more or
  42. 1:09less anything you would do with Claude
  43. 1:10Code, um you know, with one of those
  44. 1:12models instead of a traditional Opus
  45. 1:14back-end.
  46. 1:15The thing to know is, you know,
  47. 1:16obviously at really high levels of
  48. 1:18quality, which I think like the Claude
  49. 1:19Opus models are getting to, even a 1%
  50. 1:22improvement in quality might mean like
  51. 1:23really, really different results. So,
  52. 1:25I'm cognizant of that. But, I think a
  53. 1:26universe probably exists where you can
  54. 1:27fire off a lot of things with like Opus
  55. 1:294.7 or another really advanced uh kind
  56. 1:31of frontier model, and then have, you
  57. 1:33know, Deep Seek, let's say, or G-line
  58. 1:354.7 do a lot of like the heavy lifting
  59. 1:37work behind the scenes, the code
  60. 1:39refactoring, and so on and so forth, uh
  61. 1:41to save you like 5 to 10x what you what
  62. 1:42you normally would have spent. So, what
  63. 1:44I have over here on the left is a
  64. 1:46actually have Deep Seek running. You can
  65. 1:47see it's doing this in the exact same
  66. 1:48Claude Code interface, as per usual. We
  67. 1:50even have like thinking blocks that are
  68. 1:52kind of coming out. Um you can actually
  69. 1:53see that it's referencing Deep Seek V4
  70. 1:55Flash via proxy, and that's going to be
  71. 1:57the approach I'm going to show you.
  72. 1:58But, um for all intents and purposes, I
  73. 2:00mean, not really much else to say. You
  74. 2:02talk to it the same way that you do any
  75. 2:04sort of like Claude Opus model. Let me
  76. 2:05show you guys how to actually set this
  77. 2:06up. It's super easy. It takes just a few
  78. 2:08minutes, and then I can show you guys
  79. 2:09some examples of a different models and
  80. 2:11how to how to work with them. So, the
  81. 2:12way that typically Claude code works is
  82. 2:14you have it set up here in a terminal.
  83. 2:16Maybe you have it in Anti-Gravity. Maybe
  84. 2:17you have it in like the Claude desktop
  85. 2:19app or something like that. And then you
  86. 2:20send it something. So, you say, "Hey."
  87. 2:22And you know, the second that you send
  88. 2:23it a request, what it'll do is it'll
  89. 2:25push it over to its API right over here,
  90. 2:28and then you'll be billed usually based
  91. 2:29off token usage directly. Or, you know,
  92. 2:31if you're smart, what you'll do is
  93. 2:32you'll get a plan like pro or maybe max
  94. 2:35or whatever. And what these do is they
  95. 2:37basically subsidize the token usage a
  96. 2:38little bit in exchange for Claude or
  97. 2:40Anthropic, the parent company, knowing
  98. 2:42that you're going to pay a subscription
  99. 2:43every single month. And that's all well
  100. 2:44and good. It's just this is a fair
  101. 2:45amount of money, right? I would much
  102. 2:47prefer to spend $0.005 per million
  103. 2:49tokens instead of $5 per million tokens.
  104. 2:52Now, what our solution is going to do is
  105. 2:54when I send something over to Claude
  106. 2:56code, instead of it just going directly
  107. 2:58to Anthropic, what we're going to do is
  108. 3:00we're going to force it to instead push
  109. 3:02to our free Claude code proxy. And uh
  110. 3:04this is going to be set up as a local
  111. 3:06server on our computer, hence why it's
  112. 3:07this localhost, you know, 8082. You can
  113. 3:09put it on whatever freaking server you
  114. 3:10want. But, what this proxy will do then
  115. 3:12is it's then going to route it to one of
  116. 3:13three possibilities. The first is
  117. 3:15Nvidia's NIM, which uh is free, you
  118. 3:17know, with an account. I think you can
  119. 3:18also pay for some. That basically
  120. 3:20leverages a bunch of Nvidia GPUs to send
  121. 3:22requests. These models aren't as good as
  122. 3:24this next one which costs some money. Uh
  123. 3:26there's OpenRouter, which uh you know,
  124. 3:27you can get models down to like 14 cents
  125. 3:29per million. That's quite a far cry away
  126. 3:31from $25 per million, right? And then
  127. 3:33finally, you have Ollama. And Ollama
  128. 3:35actually uses like your your own GPU.
  129. 3:37So, you might have a GPU rig set up or I
  130. 3:39don't know, maybe you just have a gaming
  131. 3:40laptop or something. You can um set up
  132. 3:42models to run directly on that
  133. 3:43infrastructure. And if you're really
  134. 3:45good, you can actually set them up to
  135. 3:46run way faster than traditional uh you
  136. 3:48know, Claude models cuz you're not
  137. 3:49competing with millions of other people
  138. 3:50to send those requests. So, what we're
  139. 3:52going to do right now is I'm going to
  140. 3:53show you guys how to set up this
  141. 3:54section, which is the free Claude code
  142. 3:55proxy, and then after I'm actually just
  143. 3:57going to run through all three of these
  144. 3:59options as well as, you know, all of the
  145. 4:00models sort of on these platforms, so
  146. 4:03you guys could see everything that you
  147. 4:04need to do in order to set them up
  148. 4:05yourself. And you don't need any
  149. 4:06technical experience setting up local
  150. 4:08models or anything like that. I'm going
  151. 4:09to I'm going to walk you guys through it
  152. 4:10and handhold you from start to finish.
  153. 4:12Okay, so everything starts with this
  154. 4:13repo right here called free-claude-code
  155. 4:15by Ali Sherer. And it looks like this
  156. 4:18has gained a ton of popularity in a very
  157. 4:20short period of time. Just scrolling
  158. 4:21down here to its star history, it was
  159. 4:23basically a nothing in February and
  160. 4:24March, and just now it's hit an
  161. 4:25inflection point. A lot of people have
  162. 4:26been sharing and stuff like that. What
  163. 4:28you guys need to know is not about this
  164. 4:29specific repo. There are a lot of very
  165. 4:30similar approaches that do basically
  166. 4:31exactly what this thing does. So,
  167. 4:33uh should you not want to use this one
  168. 4:35or want to use a different one or should
  169. 4:36this become unavailable at any point in
  170. 4:38time, like just you know, don't The
  171. 4:40point The purpose of this guide is that
  172. 4:41you guys should see it in practice and
  173. 4:43then know how to set it up in your end.
  174. 4:47Okay, so what we need is we just need
  175. 4:49this little quick start down here, and
  176. 4:50I'll walk you guys through everything.
  177. 4:52Um just open up a terminal. And so, this
  178. 4:54terminal instance is currently running,
  179. 4:55you know, deepseek-v4-flash. This is
  180. 4:57like the one that's actually live. Uh
  181. 4:58so, what I'm going to do is I'm just
  182. 4:59going to exit out of this.
  183. 5:01And then I'm just going to clear. Okay?
  184. 5:02If you don't know how to do that, let's
  185. 5:04see if you have another terminal window
  186. 5:05open. You can also just head over to
  187. 5:06shell, new tab. Uh if you're running uh
  188. 5:08Windows, then you're going to have to
  189. 5:09use like Windows-specific thing called
  190. 5:11PowerShell, which is a little bit
  191. 5:12different, but um this quick start guide
  192. 5:14is going to run you guys through how to
  193. 5:15do everything regardless. So, because
  194. 5:17I'm on macOS or Linux, all I'm going to
  195. 5:18do is I'm just going to copy these three
  196. 5:20commands in.
  197. 5:21It's now going to go through and then
  198. 5:22install everything that it needs to. In
  199. 5:24my case, I already have all of it
  200. 5:25installed cuz I've obviously set this up
  201. 5:26before. Okay, then what I'm going to do
  202. 5:28is I'll just copy this next command
  203. 5:29over, give that a paste. So, it's going
  204. 5:31to clone directly in. Now, uh when I say
  205. 5:33clone directly in, you'll see this has
  206. 5:35actually changed, and that's because one
  207. 5:36of the three commands here was to go
  208. 5:38into the directory, and so now we're
  209. 5:39actually in the directory.
  210. 5:41Uh if you're on PowerShell, you just
  211. 5:42copy this command over. Um same thing
  212. 5:44with that.
  213. 5:45Okay, and so from here on out, we
  214. 5:46actually have like the folder set up.
  215. 5:48Like I know it sounds interesting and
  216. 5:49weird, but we've we've done like 80% of
  217. 5:51the work. Um just by copying this
  218. 5:53configuration, we have more or less
  219. 5:54everything we need aside from just the
  220. 5:56API key of the provider that we're
  221. 5:58using.
  222. 5:59The simplest and easiest provider to get
  223. 6:01set up with is OpenRouter, so that's
  224. 6:02what I'm going to show you guys first.
  225. 6:04And then after that we'll chat NVIDIA
  226. 6:05NIM and then finally Ollama. Um
  227. 6:07OpenRouter's probably the simplest like
  228. 6:08plug-and-play drop-in replacement for
  229. 6:11whatever you guys are currently using.
  230. 6:12If you guys have any experience sending
  231. 6:13API requests or whatever you guys for uh
  232. 6:15this video, then that'll be very very
  233. 6:17simple to you guys. And then if not, you
  234. 6:19know, I'll show you guys how to use even
  235. 6:20cheaper ones. But basically what you
  236. 6:21need to do is you just need to go and
  237. 6:22grab your your your uh OpenRouter uh uh
  238. 6:25path. So, that's pretty easy. All I need
  239. 6:27to do is just go to openrouter.ai. Now,
  240. 6:29I actually already have an account set
  241. 6:30up, so I'm just going to open up my
  242. 6:32account. And then uh here are all my
  243. 6:34requests that I've been making. I'm just
  244. 6:36going to go to API keys, and then I'm
  245. 6:38just going to go create. I'll say for
  246. 6:40YouTube.
  247. 6:41Then, I don't know, I'm just going to
  248. 6:43expire this in like a day. That way, if
  249. 6:44I forget the API key, nobody can uh you
  250. 6:47know, bill or run me up a million
  251. 6:49dollars. I'm also going to copy it, and
  252. 6:50you have to copy right there, otherwise
  253. 6:52it's going to uh basically exit out.
  254. 6:54And then what I'm going to do, and uh
  255. 6:56this may be kind of intimidating if
  256. 6:57you've never done it before, but I'm
  257. 6:58just going to find that folder. So,
  258. 6:59free-cloud-code. And the way that uh you
  259. 7:01can do this really quickly here is just
  260. 7:02PWD. So, in my case,
  261. 7:04users/nicksurvive/hello-world-claude/free-cloud-code.
  262. 7:06That's pretty straightforward. So, I'm
  263. 7:07going to go to finder.
  264. 7:09And uh I already had this open earlier,
  265. 7:11so I'm going to go hello-world-claude.
  266. 7:14And once you're done with that, you'll
  267. 7:16see the folder. Um what you guys will
  268. 7:18probably notice right off the bat if you
  269. 7:19just have like a vanilla Mac install on
  270. 7:21our developers is these files here are
  271. 7:23sort of like um you know, significantly
  272. 7:25less opaque. They're sort of shadowed
  273. 7:27out. And you may have never seen files
  274. 7:28like that before. What these are, these
  275. 7:29are hidden files. Anything on a Mac with
  276. 7:31a period in front of the file name
  277. 7:33basically just defaults to hidden. This
  278. 7:34is a convention that they use just so
  279. 7:36that, you know, if developers build
  280. 7:38something cool and and maybe potentially
  281. 7:39a little dangerous, um average, you
  282. 7:41know, computer users don't really have
  283. 7:43to worry about that. Um, but what's
  284. 7:44really cool is you can actually just
  285. 7:45show or hide hidden files at any time.
  286. 7:47And I've totally forgotten the hot key,
  287. 7:48so I'm just going to look up like macOS
  288. 7:50hidden file hot key.
  289. 7:53And, uh, right over here you can see
  290. 7:54it's command shift period. So, I'm going
  291. 7:56to go back over here, I hold command
  292. 7:57shift, press period, and boom, it just
  293. 7:59like removed all the hidden files, and
  294. 8:01then it's showing them. It's removing
  295. 8:02them, it's showing them. So, yeah, if
  296. 8:04you guys ever wonder about how to do
  297. 8:05that, um, just feel free to to look that
  298. 8:06up and it'll walk you through it. And
  299. 8:08the file you want to edit here is this
  300. 8:10on.env file right here.
  301. 8:12Okay, as you see, if I zoom way in on
  302. 8:14this, uh, once you open this up,
  303. 8:16there'll be placeholders for all of the
  304. 8:17API keys. Open router config, deepseek
  305. 8:19config, you know, LM studio config, and
  306. 8:22so on and so forth. Okay, so I'm going
  307. 8:24to paste the key in right over here, and
  308. 8:26then there's one more thing that you
  309. 8:27have to do. If you scroll all the way
  310. 8:29down to where it says open router, you
  311. 8:30also have to specify a model. And, uh,
  312. 8:32you might be wondering, where do I do
  313. 8:33that? If you scroll down, there's a
  314. 8:34little section where it says model,
  315. 8:36okay, right over here. What you have to
  316. 8:37do is you have to you have to choose the
  317. 8:39provider from open router. So,
  318. 8:40basically, if we want to use, um,
  319. 8:42deepseek four or something like that, we
  320. 8:43just go to browse all models.
  321. 8:45And then I'm just going to type deepseek
  322. 8:48v4 flash, this one right over here.
  323. 8:50There'll be this little copy right over
  324. 8:51here, deepseek {slash} deepseek, um, v4
  325. 8:54flash, okay. Once you pasted that and go
  326. 8:56open_router
  327. 8:58like this, and this should be the
  328. 9:00format, um, I've done this a couple of
  329. 9:03times. You see here it's open_router,
  330. 9:05that's where the provider, then deepseek
  331. 9:07deepseek v4 flash.
  332. 9:09If not, we'll figure it out. Okay, I'm
  333. 9:10going to save this, and once I'm done
  334. 9:12with that, um, I can actually now open
  335. 9:13up a terminal right over here, verify
  336. 9:15that I'm inside of this directory, okay,
  337. 9:17and we just did the editing on the, um,
  338. 9:19.env file, which is hidden, so you need
  339. 9:21to go ls -a if you want to see it.
  340. 9:24And then from here, we can actually just
  341. 9:25scroll up to step three, which is start
  342. 9:27the proxy, paste this in,
  343. 9:29and I may have this port set up. No, I
  344. 9:31don't, which is nice, cool. And now this
  345. 9:32is running. And now we just need to open
  346. 9:34up another terminal window. So, a lot of
  347. 9:36terminal windows here, let's just open
  348. 9:37that up.
  349. 9:38We should zoom way in, like I did a
  350. 9:40moment ago.
  351. 9:41I think that's about equal, right? This
  352. 9:43one's a little bigger.
  353. 9:44And then, uh we just run Claude Code
  354. 9:46with this proxy setup, okay? So, I'm
  355. 9:48going to copy this bash command, paste
  356. 9:51this in, and then, go enter.
  357. 9:54You can actually see we're running
  358. 9:55DeepSeek-v4-Flash via proxy down here,
  359. 9:58because we just made that request, and
  360. 9:59we sort of did that specification. Now,
  361. 10:01I'm going to go, "Hello."
  362. 10:02So, when you say, "Hello," um basically,
  363. 10:04this right over here is just a real-time
  364. 10:06log, where it's catching every incoming
  365. 10:08request. It's then sending that incoming
  366. 10:11request to, you know, DeepSeek or some
  367. 10:13other provider. DeepSeek is then
  368. 10:14catching that, and then they're sending
  369. 10:16it back. And so, you know, on the first
  370. 10:18initialization, it may take a little
  371. 10:19while, and that's more or less what
  372. 10:20happened here. The user just said,
  373. 10:21"Hello." Simple greeting. I should
  374. 10:22respond in kind with a concise, friendly
  375. 10:24greeting. "Hello, how can I help you
  376. 10:25today?"
  377. 10:26Now, what's really interesting is, when
  378. 10:27you ask it, "What model are you?"
  379. 10:30Because there's so much Claude stuff, um
  380. 10:33sort of baked into the the system
  381. 10:35prompt, it doesn't always know. And so,
  382. 10:37it's saying it's Claude Opus 4.6, the 1M
  383. 10:39context. When we're running as Claude
  384. 10:40Code, the CLI agentic interface. Um but
  385. 10:42what's really interesting is, despite
  386. 10:43the fact that it's really confident that
  387. 10:45it is,
  388. 10:46you know, if you go back to your Open
  389. 10:47Router, uh sorry, you got to go down to
  390. 10:49logs,
  391. 10:50you can actually see the request that we
  392. 10:52just sent. It's 5:39. I just sent a
  393. 10:53request over here to DeepSeek-v4-Flash.
  394. 10:56>> [gasps and sighs]
  395. 10:56>> And uh you know, you can actually see
  396. 10:58that the raw JSON down here with said
  397. 11:00request, as well. Like, this is the same
  398. 11:02request that we just sent a moment ago.
  399. 11:05And what's cool is, I mean, this one
  400. 11:06here,
  401. 11:07uh let's see, the input token was like
  402. 11:0930,000, so that's the whole system
  403. 11:10prompt of Claude Code. And then, the
  404. 11:12output was 47, which was literally like,
  405. 11:13"Hello, how can I help you today?"
  406. 11:15Uh so, we we do, unfortunately, have to
  407. 11:16feed in the whole system prompt to
  408. 11:17Claude Code. But I mean, that is
  409. 11:18literally what Claude Code is, right?
  410. 11:19It's just part of the deal.
  411. 11:21But yeah, now you literally have, like,
  412. 11:22an another model set up here. We're
  413. 11:24using DeepSeek-v4. Awesome. Build me a
  414. 11:28simple habit tracking app in a
  415. 11:32directory, let's call subdirectory
  416. 11:34called habit tracker.
  417. 11:37Make it super straightforward local.
  418. 11:40This is a demo.
  419. 11:41And I just want to give you guys like an
  420. 11:42actual live play-by-play. Because, you
  421. 11:44know, I think local models, well, not
  422. 11:46local models, sorry, non-frontier
  423. 11:47models. So, like DeepSeek instead of,
  424. 11:49you know, one of the big players,
  425. 11:50Anthropic, Google, you know, OpenAI.
  426. 11:54You know, like it's not as good and it's
  427. 11:55probably not going to be as fast. You
  428. 11:57can actually see the number of tokens
  429. 11:58that it produces over a time period.
  430. 12:00You know, like represented down over
  431. 12:02here and you'll see the speeds vary from
  432. 12:03time to time. This one over here was 30
  433. 12:0523.6, this one was 59.4. So, like, you
  434. 12:08know, it actually does take a little
  435. 12:10longer to do some things.
  436. 12:11Counterintuitively, it takes a lot less
  437. 12:13time to do other things cuz it's like
  438. 12:14different architecture, right? These
  439. 12:16models, and you know, DeepSeek in
  440. 12:17particular, I think is is out of China.
  441. 12:19These ones are like built with very
  442. 12:20different architecture in mind and they
  443. 12:22work a little bit differently. So, it's
  444. 12:23just worth it to like actually see
  445. 12:24what's going on and that's what I wanted
  446. 12:25to do here.
  447. 12:27What's really cool though is the whole
  448. 12:27Claude interface works basically the
  449. 12:29exact same as You hear how we have like
  450. 12:30the transfiguring? We got the cute
  451. 12:32little messages. It's clearly doing some
  452. 12:34thinking.
  453. 12:35You know, we can press option-enter to
  454. 12:37send a multi-line message. We can do a
  455. 12:38lot here. And now you can see that we
  456. 12:40actually have the app. So, the app lives
  457. 12:42at nicksari.com/habittracker. So, what
  458. 12:43I'm actually going to do is I'll just
  459. 12:44say open it in Chrome.
  460. 12:47And let's see if it can it can get that.
  461. 12:50So, just hit open it in Chrome and then
  462. 12:51boom, we have it right over here.
  463. 12:52>> [sighs]
  464. 12:53>> Hey, what's up? So, I'm going to add it
  465. 12:55and now I'm just going to kind of check
  466. 12:57my little habits. Boom, just did it.
  467. 12:59Fantastic and 07. Because of this, there
  468. 13:01are other strategies that you should use
  469. 13:02as well. Like, for instance, a common
  470. 13:04strategy that I like using is I just
  471. 13:06create a new instance every 50,000
  472. 13:08tokens or so because I find that quality
  473. 13:10goes down every 50,000 or so tokens. Uh
  474. 13:13that, you know, it's just not as smart
  475. 13:15or as capable. Obviously, you're trading
  476. 13:16off that intelligence and capability for
  477. 13:18money. So, what I did here is I just
  478. 13:19opened up a new uh you know, instance
  479. 13:21and then I said, "Hey, here's a file.
  480. 13:23This is way too simple. Make it look
  481. 13:24sci-quality at a serif font. Make it a
  482. 13:26much higher brow feel. Uh now it's
  483. 13:28running. So, as it correctly identified,
  484. 13:30the JavaScript is pretty clean there. Uh
  485. 13:32so, what it's going to do instead is
  486. 13:33just uh redesign both the HTML and the
  487. 13:35CSS, which is pretty neat. You can
  488. 13:37actually see the updates occurring
  489. 13:38basically live. I'm just refreshing this
  490. 13:40over and over and over again. Uh you
  491. 13:42know, it's already correctly established
  492. 13:44a nicer and sexier font, uh which is
  493. 13:47nice. So, I don't know. I mean, this
  494. 13:48isn't like an actual habit, right? So,
  495. 13:49what am I going to put here? I'll say,
  496. 13:51"20 push-ups." You know, and here I have
  497. 13:53my little thing. Nice colors, too. And
  498. 13:56you know, after another 5 or 10 minutes,
  499. 13:57I'm sure you could see how you could
  500. 13:58build something like this just as a
  501. 14:00demo. I should note that I did this
  502. 14:02entire build in like entirely using
  503. 14:03DeepSeek, whereas uh as I mentioned, I
  504. 14:06think probably the best thing to do if
  505. 14:07you really wanted a good combination of
  506. 14:08both cost and effectiveness would be to
  507. 14:10use a smarter model as an orchestrator
  508. 14:12and then have it send requests to
  509. 14:13smaller DeepSeek models with more
  510. 14:15limited context, like let's say 50 or
  511. 14:16100k. I'm just to ensure that like
  512. 14:18everybody's doing what everybody knows
  513. 14:20how to do the best. Okay, so that was
  514. 14:22OpenRouter. Let me show you guys now how
  515. 14:23to do NVIDIA NIM. Uh it's pretty
  516. 14:25straightforward, same idea. Just scroll
  517. 14:26down here to where you actually have
  518. 14:28your uh NVIDIA NIM API key section,
  519. 14:30model section. And then what you have to
  520. 14:31do is actually have to go sign up to
  521. 14:33NVIDIA's NIM platform. Now, in my case,
  522. 14:36I think I'm probably already signed up.
  523. 14:38May May May already have me signed in
  524. 14:40here. So, let me see. Yeah, it did. Um
  525. 14:42what I'll do, just to show you guys sort
  526. 14:43of what that process looks like, is just
  527. 14:44sign out. Okay, now all I have to do is
  528. 14:46sign in with my email address, which I
  529. 14:48should be able to do here. The sign-up
  530. 14:50flow is probably going to ask you to put
  531. 14:51your phone number in your email address,
  532. 14:53all that stuff. Um I'm just going to
  533. 14:54assume that you guys have phone numbers
  534. 14:55and are willing to do that. I also just
  535. 14:57don't want to share mine. And then from
  536. 14:58here, just click generate API key, and
  537. 15:00then I'll say "For YouTube."
  538. 15:02Okay, and then the expiration here,
  539. 15:04again, I'm just going to do I don't
  540. 15:05know. Why don't we do 12 hours?
  541. 15:08Then click generate key, and now you
  542. 15:09have the actual API key right here,
  543. 15:10which is awesome. Next thing you got to
  544. 15:11do is go to models really quickly. Um
  545. 15:13when you click on models, assuming that
  546. 15:14NVIDIA actually loads. Holy moly, that's
  547. 15:17slow. Um anyway, it'll open up this
  548. 15:18little models page, and then you can
  549. 15:19actually select whatever specific one
  550. 15:21you want.
  551. 15:22So, as you can see, you know, DeepSeek
  552. 15:24over here, so is GLM 5.1, GLM 4.7 right
  553. 15:28now is free, so it doesn't even cost you
  554. 15:29any money. That's what I'm going to be
  555. 15:31using just because I think that's
  556. 15:33probably like the simplest one
  557. 15:35to get set up with. And then, uh, just
  558. 15:37because this is taking a while, I'm just
  559. 15:39going to go Z A I / GLM
  560. 15:41-4.7. Okay, now we're actually in a
  561. 15:44little test page and we can actually
  562. 15:45have a conversation with it. We can
  563. 15:46actually say, "Hey."
  564. 15:48Um, this is now going to send the
  565. 15:49request to the model. The model's going
  566. 15:50to go through its whole thinking,
  567. 15:52uh,
  568. 15:53whatever, and then, you know, so I guess
  569. 15:55we just verify that that works. Not
  570. 15:56super relevant for us. All we have to do
  571. 15:58now is if you go back to this quick
  572. 16:00start, you'll see that it says
  573. 16:00nvidia_nim/z-ai/glm4.7.
  574. 16:04So, I think we actually probably already
  575. 16:05have this set up. I'm just going to
  576. 16:06double-check. I just go back to finder
  577. 16:08where you had that .env. Okay. Then from
  578. 16:11here, um, just scroll down until you
  579. 16:13have the NVIDIA API key, which we do.
  580. 16:15So, I'm just going to paste in my NVIDIA
  581. 16:17API key. Zoom in a bit for you guys.
  582. 16:19Then scroll down to again where it says
  583. 16:21model, and this is going to be where,
  584. 16:23uh, you'll put in the NVIDIA one, the
  585. 16:25one that we just had a moment ago. So, I
  586. 16:26think that's probably the one that we'll
  587. 16:28do. I'll use NVIDIA NIMs GLM 4.7 because
  588. 16:32it's free, and, you know, I want to show
  589. 16:33you guys how to do a free model.
  590. 16:35We'll paste that in. And now I'm just
  591. 16:36going to hold command and then press S.
  592. 16:39That's going to save it. Uh, you can you
  593. 16:41can see there are a lot of other, you
  594. 16:42know, different types of models and so
  595. 16:43on and so forth. And now what we can do,
  596. 16:45you know, this is currently actually
  597. 16:47doing a build right now, um, but as you
  598. 16:48can see, the tokens have stalled. And
  599. 16:50the reason why the tokens have stalled
  600. 16:51is cuz I just like changed the the file.
  601. 16:53What I What I'm going to do is I'm just
  602. 16:54going to delete both of these, okay? And
  603. 16:56I'm doing that by holding control and
  604. 16:57then pressing, uh, C.
  605. 17:00I'm going to clear these out.
  606. 17:01>> [sighs]
  607. 17:01>> And now I'm just going to run the server
  608. 17:03again. That's 8082. And then over here,
  609. 17:06I'm just going to go Anthropic off
  610. 17:07token, Anthropic base URL
  611. 17:08localhost:8082/cloud.
  612. 17:11Okay, so now I'm going to go, "Hey."
  613. 17:13It looks like the reason we're getting
  614. 17:15this is speed. I think that's because I
  615. 17:17have fast mode on, so let me just turn
  616. 17:18that off.
  617. 17:20Okay, cool. So, we just went hey, and
  618. 17:21then it's sending a request over to
  619. 17:22Nvidia now.
  620. 17:25So, now it's saying, "Hey, how can I
  621. 17:26help you today?" Fantastic. What you'll
  622. 17:28find is if you're using things that
  623. 17:30aren't supported by these models, like
  624. 17:31for instance, these models don't have
  625. 17:32fast mode, even though that's sort of
  626. 17:33like enabled within Claude Code, you
  627. 17:35send the request, you'll receive an API
  628. 17:37error being like, "Hey, what the hell's
  629. 17:38a speed thing? I don't have that." So, I
  630. 17:40just make sure you turn any of that any
  631. 17:42and all of that stuff off.
  632. 17:44Uh and then I'll say, "Hey, what's your
  633. 17:45working directory?" Just cuz I want to
  634. 17:47double-check and see what it's at.
  635. 17:50And uh you'll also notice that this is
  636. 17:51still broken. It says deep sea before
  637. 17:53flash, which is kind of funny. Uh this
  638. 17:55one says hello world Claude free Claude
  639. 17:57Code, awesome.
  640. 17:59Um you know,
  641. 18:02"How do we change the status line?"
  642. 18:05Let's just ask it that. Maybe you can
  643. 18:07actually do that. So, they can see here
  644. 18:09that that response took a little while,
  645. 18:10and then it asked about changing the
  646. 18:12status line. Um this thinking is
  647. 18:14occurring here in just one giant block
  648. 18:16as opposed to streaming, which is
  649. 18:17interesting. So, what it's doing now is
  650. 18:19it's spawning the status line setup
  651. 18:21agent, okay, which is something that's
  652. 18:22actually built into Claude Code,
  653. 18:24interestingly enough.
  654. 18:25But again, we're using a different model
  655. 18:27brain
  656. 18:28to power it. And so, that's what's
  657. 18:30really cool about this harness, like
  658. 18:31it's interoperable, and you can swap
  659. 18:32whatever other agents are inside of this
  660. 18:34container. Okay, and here it's now
  661. 18:35asking me what the model name is. For
  662. 18:37model name, I'm just going to say use
  663. 18:38GLM 4.7 for model name, and then
  664. 18:42everything else the same.
  665. 18:44And let's see if it can actually go do
  666. 18:45it, if it has enough context and
  667. 18:47intelligence to do it natively on its
  668. 18:48own. Okay, and then the last thing we
  669. 18:50need to do is we need to set up Ollama.
  670. 18:51So, Ollama's actually pretty
  671. 18:52straightforward. Just head over to
  672. 18:53ollama.com, and then there'll actually
  673. 18:55be a one-liner that you can just paste
  674. 18:56into your terminal. You can also
  675. 18:58download it through like more
  676. 18:59conventional means, um but I'm just
  677. 19:00going to use the shell, cuz that's easy
  678. 19:02for me.
  679. 19:03These guys are pretty cute. They're a
  680. 19:04lot more visual, so all their stuff will
  681. 19:05be really adorable. Once you're done,
  682. 19:07just type in Ollama,
  683. 19:09and it'll actually like get Ollama up
  684. 19:11and running on your computer. So, as you
  685. 19:12can see here, there's now chat with a
  686. 19:13model, launch open claw, launch claw to
  687. 19:15code, launch open code more, right?
  688. 19:17There's a lot more down here. Whatever
  689. 19:19files you want. Um just for examples,
  690. 19:21I'll go chat with a model, and then what
  691. 19:23do we want? I'm going to really simple
  692. 19:24one.
  693. 19:25Uh you can see that a couple of these
  694. 19:26are cloud, right? So, this one's
  695. 19:27actually in the cloud. That one's in the
  696. 19:28cloud. That one's in the cloud. That
  697. 19:29one's in the cloud.
  698. 19:30But, I want something like really,
  699. 19:31really, really basic and really simple.
  700. 19:33I think I'll probably just use Gemma 4
  701. 19:35because this one's 12 gigs, and I can
  702. 19:37probably download it through the video.
  703. 19:38So, I'll just click Gemma 4, download
  704. 19:39Gemma 4.
  705. 19:40Now, it's going to pull the manifest,
  706. 19:41and actually have to do the the
  707. 19:42download. Now, my internet's pretty
  708. 19:43fast, so I should be able to download,
  709. 19:45you know, this whole thing, which says
  710. 19:47it's around 10 GB or so in a couple of
  711. 19:49minutes. I'm just going to circle back
  712. 19:50when it's done. We'll also go back here
  713. 19:52to models, and then what we wanted was
  714. 19:56Oh jeez, what did we just download? I
  715. 19:58think it was Okay, awesome. So, we
  716. 20:00actually have Gemma 4. And the reason
  717. 20:01why I'm doing this is because I just
  718. 20:02want to get the uh the code. And it
  719. 20:03looks like the code is just Gemma 4
  720. 20:05itself, all lowercase. So, we should be
  721. 20:06able to do that pretty straightforwardly
  722. 20:08when it's done. K, and then I think
  723. 20:11we're now actually talking to it. So,
  724. 20:12I'll just say, "Hey."
  725. 20:14And yeah, you can see the thinking
  726. 20:15process is now being written. Uh what's
  727. 20:17cool is this is now entirely local on my
  728. 20:18computer. Um are you
  729. 20:21on the cloud or what? I don't actually
  730. 20:23know if it's going to know.
  731. 20:24>> [gasps]
  732. 20:25>> It's asking a philosophical question.
  733. 20:26Yeah, no, it clearly doesn't actually
  734. 20:28know where it's running. Um that's
  735. 20:29pretty funny. Anyway, regardless, uh we
  736. 20:31have basically everything that we we
  737. 20:32need set up, which is cool. So, I'm just
  738. 20:34actually just going to exit out of this.
  739. 20:35I think it's control D.
  740. 20:37K, once you're done with this, I'll just
  741. 20:38click control C. [snorts]
  742. 20:40K, and I'll clear it. Okay, and then
  743. 20:42once we're done, um we just need to
  744. 20:43write Ollama pull, and then it'll pull
  745. 20:45Gemma 4.
  746. 20:47That will actually like go and grab
  747. 20:48Gemma 4. I mean, in my case, I already
  748. 20:50have it on my computer cuz we just
  749. 20:51downloaded it. But, uh from there, we
  750. 20:52just go Ollama serve. Uh and I think the
  751. 20:55reason why it's served is it I think it
  752. 20:57like auto serves um when you do that
  753. 20:59now. It might not have used to do that.
  754. 21:01Alternatively, we might have just
  755. 21:02already been running it, which is great.
  756. 21:04Um that means we can now go through and
  757. 21:05then update the last thing that needs
  758. 21:07updating and then open up the file and
  759. 21:09then look for Ollama. You can see down
  760. 21:11here it says Ollama config local
  761. 21:12provider no API key provided. Then it
  762. 21:14gives me the base URL. And if you guys
  763. 21:16notice the base URL localhost colon
  764. 21:1811434 is the same as the 11434 here. And
  765. 21:21the localhost, another way to say that
  766. 21:23is just 12700
  767. 21:25one. So what this is saying is like,
  768. 21:26"Hey, you're currently running the
  769. 21:28server for Ollama on this."
  770. 21:30Which is obviously awesome to us. Okay,
  771. 21:31and then the last thing we have to do is
  772. 21:32we have to actually grab
  773. 21:34um whatever this is, Gemma 4 latest.
  774. 21:37Okay, and then copy that over and then
  775. 21:39paste it in right here.
  776. 21:41And we always have to prepend with
  777. 21:42Ollama if it's if it's listed up here.
  778. 21:44So Ollama and then Gemma 4 colon latest.
  779. 21:47Save that and then now we can actually
  780. 21:48go and run the app again. So we just ran
  781. 21:50the app and then uh I think it's over
  782. 21:53here.
  783. 21:54So we connect to that. And now we should
  784. 21:56be having a conversation with um Gemma
  785. 21:594. So I'll say, "Hey." Let's see how it
  786. 22:00goes.
  787. 22:01You can see this is sent I think to a
  788. 22:03different endpoint. So this tells me
  789. 22:05it's probably right. What's really
  790. 22:06interesting is I just sent that and my
  791. 22:08computer just started getting really
  792. 22:10hot. The reason why is because all of
  793. 22:11this was occurring locally on my on my
  794. 22:14own computer. I can actually hear it
  795. 22:15humming now, which is really funny.
  796. 22:18Um that's because it's literally passing
  797. 22:20through that request and doing all the
  798. 22:21transformer multiplication uh matrix you
  799. 22:24know inversions or whatever the hell,
  800. 22:25like literally on my own computer, which
  801. 22:28is really cool to me.
  802. 22:29And it's now dying down because it's not
  803. 22:32actually doing the the run anymore.
  804. 22:34So that took a fair amount of time just
  805. 22:35to get this response, right? The reason
  806. 22:37why is because when you run on local
  807. 22:38models like Ollama, um it's it's really
  808. 22:40slow like to be honest because it's
  809. 22:41running on your own hardware, right? Uh
  810. 22:43it's differently if you're running on
  811. 22:44like dedicated GPUs like with OpenRouter
  812. 22:46or um even like Invidia NM.
  813. 22:48But here like I I am literally running
  814. 22:49on my MacBook and my MacBook is not
  815. 22:51optimized for this at all. So hopefully
  816. 22:52I'm still capable of recording without
  817. 22:54uh skipping frames or whatever, but
  818. 22:56yeah, that's uh that's how you run it
  819. 22:57locally on your on your computer as
  820. 22:58well.
  821. 22:59And I should note that this is a big
  822. 23:00model, you know? it's like uh you know,
  823. 23:0210 GB with 128k contacts running
  824. 23:05locally. You know, if you don't have a
  825. 23:06pretty good computer, you'll probably
  826. 23:08have to use a dumber model. So, um I
  827. 23:10don't know, you can go newest here to
  828. 23:11sort by like the the good ones, but a
  829. 23:13pretty reliable one was the Llama series
  830. 23:15of models for a while. So, uh you know,
  831. 23:17if you're running something a little bit
  832. 23:18smaller, maybe a Llama Let me see, Llama
  833. 23:203.1, 3.2.
  834. 23:23These ones are okay.
  835. 23:24Uh this guy said it was Mythos, which is
  836. 23:26hilarious.
  837. 23:28You'll also see that uh models on Ollama
  838. 23:30are from different people.
  839. 23:32This is by Siddiqui
  840. 23:33uh Siddiqui Nashra, which is pretty
  841. 23:35interesting. And so, typically what
  842. 23:36they'll do is they'll like modify the
  843. 23:38model a little bit, they'll change it,
  844. 23:40they'll fine-tune it, they'll just make
  845. 23:41it make it different. So, feel free to
  846. 23:43take a look see through Ollama's um
  847. 23:46database of models and see if there's
  848. 23:47anything maybe that like works for you.
  849. 23:49A payroll-specific model,
  850. 23:50accounting-specific model, and and what
  851. 23:52have you. One more thing that I'll
  852. 23:53mention is um earlier I talked about
  853. 23:56having Claude orchestrate conversations
  854. 23:58between Deep Seek and then you know, a
  855. 24:00smarter model in order to do stuff. And
  856. 24:02so, what I've done here, if I just get
  857. 24:03my ugly head out of the way,
  858. 24:05is um is I've done that. I've uh
  859. 24:07basically said, "Hey, send a message to
  860. 24:09Deep Seek using the flow we created
  861. 24:10earlier, then tell me what it says
  862. 24:12back." And so, that's what it's doing
  863. 24:14right now. It's actually basically doing
  864. 24:15all this hosting stuff totally on its
  865. 24:17own.
  866. 24:18And as you can see, when you just send a
  867. 24:19message directly to Deep Seek, it says,
  868. 24:20"I am Deep Seek, the latest version of
  869. 24:22the AI model developed by Deep Seek
  870. 24:23company."
  871. 24:24Um you know, okay, great. Orchestrate it
  872. 24:26to build an app, guide it, and suggest
  873. 24:29it
  874. 24:30through the steps. App description,
  875. 24:34I don't know, simple calorie tracker,
  876. 24:37but MVP needs to appear super luxe and
  877. 24:40high-end for the purposes of a demo.
  878. 24:44And um you know, what's happening right
  879. 24:45now is like my much smarter model, Opus
  880. 24:484.6 in this case. I know it's saying GLM
  881. 24:504.7 down here, but that's because we
  882. 24:51actually just updated the status line
  883. 24:53manually, which you shouldn't probably
  884. 24:54do, by By way. I I hardcoded in GLM 4.7
  885. 24:57just as proof of concept.
  886. 24:59Uh but we're doing is we're using this
  887. 25:00much smarter Opus model to basically
  888. 25:01communicate with DeepSeek through the
  889. 25:04proxy server that we just set up. So,
  890. 25:06it's actually using, you know, Claude
  891. 25:08code under the hood, and then we're
  892. 25:08going back and forth. And as you can
  893. 25:10see, our simple prompt here was
  894. 25:13understood, digested, and then spat out
  895. 25:15by Opus. So, it's a lot more complex,
  896. 25:17and actually has some opinionated
  897. 25:18baked-in font choices and
  898. 25:20colors and stuff like that. And um it's
  899. 25:22just sending all of this over to
  900. 25:24DeepSeek right now. And it's just going
  901. 25:26to go back and forth with DeepSeek until
  902. 25:27we have, you know, an okay-looking app.
  903. 25:29And as you guys can see, the
  904. 25:30orchestration sort of approach put
  905. 25:31together this app, which is kind of
  906. 25:32neat. Uh this is just a a base calorie
  907. 25:34tracker, but still looks pretty cool.
  908. 25:36Obviously, some minor issues here, like
  909. 25:38the spacing. I think if I build in a
  910. 25:40better orchestration loop, like I had it
  911. 25:42do screenshots and stuff, it probably
  912. 25:43would have sorted that out. It even has
  913. 25:45some like quick-add calorie stuff, which
  914. 25:47is wild. Then you actually have the the
  915. 25:48log here as well. So, you can get, you
  916. 25:50know, arbitrarily complex, more or less
  917. 25:52as complex as you want. Looks like it
  918. 25:53even did a little calorie tracker app,
  919. 25:55which is awesome.
  920. 25:56So, that's pretty cool. And um you know,
  921. 25:58when Anthropic ran the same sort of
  922. 26:00infrastructure, uh I think pairing Opus
  923. 26:024.6 with I think Sonnet 4.6, or maybe it
  924. 26:04was Opus 4.5 with Sonnet 4.5 sub-agents,
  925. 26:07cuz this is basically what we're doing,
  926. 26:08a sub-agent flow. Uh I think it
  927. 26:09performed like, I don't know, 15% better
  928. 26:12or something like that.
  929. 26:13So, what we're doing now is we're not
  930. 26:14just using sub-agents made by the same
  931. 26:16company. We're using sub-agents that are
  932. 26:17hundreds of times cheaper to do a lot of
  933. 26:19our intellectual heavy lifting. And then
  934. 26:20the Opus series is just going to take a
  935. 26:22look at it and be like, "That's good."
  936. 26:23or "That's bad." which is obviously a
  937. 26:25lot more profitable. Okay, so hopefully
  938. 26:26you guys appreciated that video. I had a
  939. 26:27lot of fun putting it together. Uh costs
  940. 26:29were a major concern of mine up until
  941. 26:30just a few years ago when I started, uh
  942. 26:32you know, making a fair amount of money
  943. 26:33with this stuff. So, obviously this sort
  944. 26:35of intelligence should be equally
  945. 26:37distributed, but for the near term, I
  946. 26:39doubt that it will be, simply because,
  947. 26:40you know, uh model companies got to eat,
  948. 26:43I guess. GPUs got to go burr.
  949. 26:45If you guys like this sort of thing, and
  950. 26:46you want to see how to maybe make some
  951. 26:48money as opposed to try and squeeze your
  952. 26:49wallet as much as you possibly can,
  953. 26:51definitely check out Maker School, it's
  954. 26:52my day-by-day accountability program
  955. 26:54where I basically show people how to
  956. 26:55take their skills with Cloud Code and
  957. 26:57then other no-code automation platforms
  958. 26:59as well like N8n and so on and so forth.
  959. 27:01And then you use that to sell AI agents
  960. 27:03and full workflow automation to
  961. 27:05companies to make a living. We have a
  962. 27:0790-day guarantee where you will get your
  963. 27:09first client within 90 days or I will
  964. 27:10pay you all of the money that you spent
  965. 27:11on my program back. It's been run
  966. 27:14through by over 10,000 people to date
  967. 27:16and I'm not going to sit here
  968. 27:17advertising any more than that but it's
  969. 27:19pretty cool. There's basically no
  970. 27:20downside to signing up. Aside from that,
  971. 27:22if you could do me a big solid, just
  972. 27:23like and subscribe to the video. Every
  973. 27:25little bit of engagement helps and I'm
  974. 27:26starting a big YouTube run here so
  975. 27:27that'd be quite valuable. And yeah, let
  976. 27:30me know what you guys want to see next
  977. 27:31because I make videos like this
  978. 27:32basically entirely based off of your
  979. 27:33requests and recommendations. This one
  980. 27:35was made directly by a request that I
  981. 27:37received a couple of days ago. Have a
  982. 27:38lovely rest of the day and I'll catch
  983. 27:40all y'all in the next video.

About this transcript

This page contains the full transcript of How to Use Claude Code for FREE (2026) by Nick Saraev, generated from the public captions YouTube serves with the video. The transcript has 6,843 words across 983 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 choose a caption track available on the video.

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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.