How to USE Claude Code for FREE with Ollama ( Local AI FULL Tutorial) — Transcript

Full transcript

1. 0:00Whether you have a high-end PC, a
2. 0:03powerful laptop running Windows 11,
3. 0:05Linux, or even a MacBook with Apple
4. 0:08silicon,
5. 0:10you can start using Claude Code with
6. 0:12Ollama's local and cloud models. For
7. 0:15this tutorial, we will be installing
8. 0:18Claude Code on Linux.
9. 0:28The first, let's talk about my hardware
10. 0:30specs. I'm running this on my HP gaming
11. 0:33laptop. It's decent, but definitely not
12. 0:36a supercomputer. I'm mentioning this
13. 0:38because AI performance depends heavily
14. 0:41on your hardware. You will need at least
15. 0:4416 GB of RAM and Nvidia GPU with 4 GB of
16. 0:48VRAM. If your system meets these
17. 0:51requirements, you can run models like
18. 0:53Qwen 3.5 and GLM 4.7 Flash locally
19. 0:58without any major issues.
20. 1:01If your laptop struggles with local
21. 1:03models, don't worry. I will share a free
22. 1:06cloud-based solution later in the video.
23. 1:13As you can see, I have also installed
24. 1:15proprietary Nvidia drivers to enable
25. 1:18partial GPU acceleration.
26. 1:23Since we are running Claude Code with
27. 1:25Ollama, the first step is to install
28. 1:28Ollama on your system.
29. 1:34Go ahead and open your web browser. Go
30. 1:37to official website, copy the install
31. 1:39command, and paste it into your
32. 1:41terminal. If you're on a Mac, simply
33. 1:45download the DMG file, drag the icon
34. 1:47into applications folder, and open
35. 1:49Ollama. It will start running in the
36. 1:52background.
37. 1:54The process is similar on Windows.
38. 2:00All right, once Ollama is installed, as
39. 2:02you can see, it has detected the Nvidia
40. 2:05GPU. This means you can run compatible
41. 2:08LLMs with full GPU support. Now, let's
42. 2:12verify that Ollama is running.
43. 2:16Enter this command to check its status.
44. 2:19You will notice that the default context
45. 2:21window is set to 4K.
46. 2:31By default, Ollama runs on a local port.
47. 2:34Open this URL in your browser to confirm
48. 2:37that Ollama server is working properly.
49. 2:44Next, it's time to install Claude Code.
50. 2:47This is an AI coding assistant that can
51. 2:49write, understand, and fix code for you.
52. 2:52You can install Claude Code based on
53. 2:55your operating system. On Linux and
54. 2:57macOS, it's just a simple one-line
55. 3:00command.
56. 3:01On Windows, you will need to run the
57. 3:03relevant command inside PowerShell.
58. 3:07I will copy this command and paste it
59. 3:09into the terminal.
60. 3:13This will start the installation
61. 3:14process.
62. 3:19Once it's done, you will need to run
63. 3:22this command to add Claude Code to your
64. 3:24system's environment variables.
65. 3:30Now, you can verify that Claude is
66. 3:32installed by running this command. As
67. 3:35you can see, it shows the version, which
68. 3:38means Claude is successfully installed.
69. 3:43You can run Claude directly to use
70. 3:45Anthropic models like Opus if you have
71. 3:48an account.
72. 3:49But for now, we will stick with Ollama.
73. 3:52To exit the TUI, press control plus C
74. 3:55twice.
75. 3:58Next, it's time to run Claude Code with
76. 4:00local Ollama models. This means you will
77. 4:03be using your own hardware to power the
78. 4:06AI.
79. 4:07The first, we need to download a model
80. 4:09that's both powerful and efficient.
81. 4:12Let's go to Ollama's model library,
82. 4:15apply the tool link and thinking
83. 4:17filters, and choose Qwen 3.5.
84. 4:21It's a great balance of performance and
85. 4:23efficiency, especially for low-end
86. 4:25systems like mine.
87. 4:28If you have more RAM and VRAM, you can
88. 4:30try models like Qwen 3 Coder and GLM
89. 4:34Flash 4.7.
90. 4:36But for most cases, Qwen 2.5 or 3.5 is
91. 4:39the great choice.
92. 4:41Now, go ahead and copy this command and
93. 4:43run this inside terminal to download the
94. 4:46model.
95. 4:50This will install the 6 billion
96. 4:52parameter version with a large context
97. 4:54window.
98. 5:01Once installed, let's test it in REPL.
99. 5:05As you can see, it runs decently on my
100. 5:07system using both CPU and partial GPU
101. 5:11acceleration.
102. 5:13Now, by default, Ollama intelligently
103. 5:16splits the workload so performance stays
104. 5:18stable.
105. 5:19Now, go ahead and press control plus C
106. 5:22to stop the model from generating the
107. 5:23output and exit the REPL environment by
108. 5:26pressing control plus D.
109. 5:29Now, by default, the context window is
110. 5:31limited, which isn't ideal for Claude
111. 5:34Code. Claude requires minimum of 64K
112. 5:37context window.
113. 5:39We will increase it by creating a custom
114. 5:42model with a larger context size.
115. 5:45Now, go ahead and create a model file
116. 5:47using this command.
117. 5:49Then set the base model and context size
118. 5:52of 64K, then save it.
119. 6:02After that, run the command to create a
120. 6:05brand new custom model with any name of
121. 6:07your choice.
122. 6:17You can verify it using Ollama list.
123. 6:22Now, go ahead and run this brand new
124. 6:24model. If it loads successfully, your
125. 6:27system can handle the large context.
126. 6:35You can also confirm by running this
127. 6:37command, and you can see it uses the
128. 6:40larger context window.
129. 6:42Now, let's connect this model to Claude
130. 6:44Code. Just go ahead and create a folder
131. 6:47for your workspace and navigate into it.
132. 6:54Then launch Claude Code using your
133. 6:57custom model.
134. 6:59Now, simply type the brand new Ollama
135. 7:01launch command with the model name.
136. 7:09You will see the Claude setup screen.
137. 7:12Just go ahead and choose your theme and
138. 7:14trust the workspace. And just like that,
139. 7:16Qwen 3.5 is now running locally with
140. 7:20Claude Code.
141. 7:27Now, let's test it with a simple prompt.
142. 7:48As expected, responses are slower since
143. 7:52everything is running locally, but it
144. 7:54works perfectly.
145. 8:01Now, for something more practical, let's
146. 8:04ask it to create a simple C program and
147. 8:07save it to a file.
148. 8:18As you can see, it successfully uses
149. 8:21tool calling to create and manage files.
150. 8:29We can even ask it to run and test the
151. 8:32program, and it works perfectly.
152. 8:53And lastly, let's try generating a shell
153. 8:56script to create multiple folders.
154. 9:19And there we go. It completes the task
155. 9:21without any issues.
156. 9:23And that's it. Now, you have Claude Code
157. 9:26running locally with Ollama.
158. 9:31Next, it's time to show you how to use
159. 9:33Claude Code with Qwen 3.5 in VS Code to
160. 9:37build a simple static website. The
161. 9:39first, create a brand new workspace and
162. 9:42open it inside VS Code.
163. 9:50Then open terminal inside VS Code and
164. 9:53launch Claude using the Qwen 3.5 model.
165. 10:01I will keep accept edits turned on. This
166. 10:05way, it can automatically apply changes.
167. 10:08Now, let's provide a prompt to create a
168. 10:10simple Hello World website using HTML
169. 10:14and CSS.
170. 10:29After a few minutes, it generates a
171. 10:31complete HTML file with all the content.
172. 10:38As you can see, this is the code it
173. 10:41wrote.
174. 10:45And here's the website it created. It
175. 10:48actually looks very good. Since I'm
176. 10:50using a low-end system, the responses
177. 10:52are a bit slower. But on higher-end
178. 10:55machines like Apple Silicon with 64 GB
179. 10:57of unified memory, the generation speed
180. 11:00will be much faster.
181. 11:02If your computer doesn't have at least
182. 11:0516 GB of RAM and 4 GB of VRAM, you may
183. 11:08not be able to run Ollama models
184. 11:11locally. In that case, you can use
185. 11:13Ollama's cloud models. They offer a free
186. 11:16tier allowing you to run powerful LLMs
187. 11:20with high-end GPUs. These are the cloud
188. 11:23models available for free. Now, for this
189. 11:25video, I will be using Kimik 8 2.5 cloud
190. 11:29edition, which works great in most
191. 11:32scenarios.
192. 11:33Just go ahead and run this simple
193. 11:35command, and it will prompt you to open
194. 11:38a web browser and sign in into your
195. 11:40Ollama account. Log in using your email
196. 11:43or Google account, then connect your
197. 11:46device.
198. 11:51Once it's done, return to the terminal.
199. 11:55Now, you can use Cloud Code with the
200. 11:57Kimik 8 2.5 cloud edition. It's very
201. 12:00good for agent-based task.
202. 12:06Now, let's test it with the prompt.
203. 12:11And as you can see, the response is
204. 12:13super fast.
205. 12:44You can exit the interface anytime using
206. 12:47the exit option.
207. 12:52Now, let's use Kimik 8 2.5 in VS Code to
208. 12:56create a simple to-do list application.
209. 13:06It completes the task in 30 to 40
210. 13:09seconds, which is insanely fast.
211. 13:15And as you can see, the website looks
212. 13:17great.
213. 13:25And that's pretty much it. This is how
214. 13:27you can run Cloud Code using Ollama
215. 13:30models.
216. 13:31Let me know what you think about this in
217. 13:33the comment section down below.
218. 13:35If you like this video, hit that like
219. 13:37button and subscribe to see more
220. 13:39content. Thank you so much for watching.
221. 13:42This has been KSK Royal. I will see you
222. 13:45in the next one.