Transcription
[Music] [Applause] Hi, welcome to another video. So, Claude Code has been a buzz recently. People use it a lot and everything. It especially comes in handy when you use it with the $200 or $100 thing because it makes it kind of unlimited, but you can also use it via APIs as well.
The major thing about Claude Code is that its UI is good. It's prompted quite well, and the tools that it uses are also quite good. But the major issue with Claude Code is that it's just bound to the Anthropic models. Meaning that you can only use the Claude models, and none of them is cheap. Plus, it is also not open source, which leaves you with the option to only use something like Gemini CLI or Codeex CLI, which are not as well-versed. And the ecosystem around them is also not good. For example, if you want a graphical interface, then Claude has that, but it doesn't work with Gemini CLI and stuff like that.
However, now there's an option to do just that with Claude Code Router. Claude Code Router is a tool for routing Claude Code requests to different models, and you can customize any request. Yes, it basically makes Claude Code connect to any model that you have, whether it be a local model via OAMA or a Gemini model or Mistral or even DeepSeek or anything like that.
Now configuring it can be a bit tricky because you actually need to configure about four models. Yes, actually Claude Code doesn't use one model. It uses a ton of them for different aspects, and you would need to set each one of them in order to use it. You need a background model, which is primarily just a model that does some stuff like haiku generation for the loading messages, conversation summarization, and command processing as well. It doesn't do anything complex. So you can either have a super small model for it running locally or you can just use some open router free model. I recommend you to use the Gemini 2.0 Flash model for this.
Then there's the think model, which basically is the reasoning model. It is enabled when you ask Claude Code to think. I use Gemini 2.5 Pro for this as well as in the general model option. That's because that model is really good, and it is now free in the API as well. Yes, it now gives you about 1,500 requests per day for free, which is pretty awesome. And you can use that API for free with thinking budgets and stuff like that, which is awesome. For the long context model, I also use the same model as well as in the model option; I also use the same model. So that is about the model configuration.
Another thing that it has is the plugins. Plugins basically modify the request to have things like more system prompts in order to make your models work well. For example, this plugin works well for DeepSeek models.
Anyway, now let me show you how you can actually use it as well. But before we do that, let me tell you about Ninja. Ninja Chat is an all-in-one AI platform where for just $11 per month, you get access to top AI models like GPT-40, Claude 3.7 Sonnet, and Gemini 2.0 Flash, all in one place. I've been using Gemini for quick research, but what's really cool is their AI playground where you can compare responses from different models side by side. Their mind map generator is a game-changer for organizing complex ideas as well. The basic plan gives you 1,000 messages, 30 images, and five videos monthly, with higher tiers available. If you need more, use my code king25 for 25% off any plan or king40yearly for 40% off annual subscriptions. Check the link in the description to try it yourself.
Now, back to the video. So, using it is simple. You'd have to first install Claude Code Router with this command, and you should be good to go. Now, first you'll need to run Claude Router with the CC start command, and that will start the router. For the first time, it will also ask you for the model that you want to use. So, the model that you configure here will be used as the main model in all places. But you can also configure different models for different things in the config. Like I said before, simply you can just enter the Gemini base URL, API key, and model name, and you should be good to go. You can change the config to have different models for each thing if you want, and that is also fine.
Anyway, that's the config, and now we can start using it. To use it, just run the CCR code command, and that will start Claude Code immediately with the router. And now it is using the Gemini model while everything remains the same. Just for your info, it works very seamlessly. If you see a bit of delay in getting responses, then that is because Gemini's free tier is slow at times, and that is why it may become slow, but with paid models, it should work even smoother.
Anyway, I can send a hi message here, and you'll see that it will go ahead here, and then in a bit it will return the message. Anyway, now let's try to use it as well. I'm going to ask it to make me a minesweeper game using HTML, CSS, and JS. This is a simple one, and I'm just showing it for a test here. I have tested it on multiple tasks, and I can say that at least with Gemini it works as well if not better because of the long context window. So anyway, it will go ahead and start to work on it. You can see that it works just like Claude would, with just the model differences and nothing else. It is really funny how good this works with Gemini. It almost seems like I'm using the general Claude Code, and the tools and everything work fine along with MCPS and everything, which is quite awesome as well. In a bit it gets done, and it did all this amazingly well. It doesn't show any errors at all and stuff. It can behave sometimes inappropriately. Like it can give errors and stuff for diff editing, but that is in very rare cases, and you can always fix that with the plugins option that this one provides as well.
I really liked it, and it is actually good and useful. You can also configure it to use with things like GitHub actions for CI workflows or stuff like that. I actually like Claude Code, but it not being open-source is a bummer, and it doesn't fix that. But at least it places users in a position to use whatever model that we want with it as well. It is really good and works well without a ton of issues, which is something that happens with a lot of alternatives, but this is the best one and actually works well.
Overall, it's pretty cool. Anyway, share your thoughts below and subscribe to the channel. You can also donate via the Super Thanks option or join the channel as well and get some perks. I'll see you in the next video. Bye. [Music]