Transcription
Hi, welcome to another video.
So, Kimmy has launched their CLI and coding plan. I saw this and thought to talk about it as well. Now, this is basically their own new CLI coding agent that is mostly made to be paired with their subscription plans. It is open source, which is great. This is just like any coding agent, but because it's from the AI model lab itself, you can think that this works much better with the nuances of the KI lineup models.
So, let's first talk about the plans that you can use with this. All Kimmy subscriptions give you a weekly limit with different quotas. For example, the lowest cost version of $20 gives you about 248 weekly queries, which if you divide by 7 gives you about 250 requests a day, which is enough for light tasks. However, this is a weekly limit. So, if someday you need like 1,000 or even 2,000, then you can get that, which is probably the best thing about this because you can manage your weekly quota accordingly. $20 is pretty affordable as well. There are also bigger plans, but the limits are not clear to me yet as they're not updated on the pricing page, but there's that. It looks like a pretty good deal on paper, but I don't know if each request is counted as one request or if a bigger request takes up more quota or something like that. I'm still yet to test that.
Now, you can use this API in Kilo, Rue, Klein, Open Code, and stuff like that as well. So, it's very similar to the GLM coding plan, but their own Kimmy CLI might be a bit better because it's fine-tuned for their models and they know the nuances of their models better than generalized tools. Their CLI is built on Python which is interesting because generally the choice has been TypeScript or Rust or something similar for CLI tools but this is based on Python and you'd have to install it with UV and stuff like that as well. This is also fully compatible with the ACP protocol by Zed. So you can integrate it into Zed and even Jet Brains because I think that also now supports the ACP protocol. More tools should start incorporating ACP support as well.
Anyway, one more interesting thing is that it's even a shell. Yes, it actually allows you to hit tab and switch between shell and agent mode. So you can basically make it your entire shell that automatically opens up when you launch the terminal and then switch between shell and agent mode instantly. I'll show that in a bit. But it also has some cool features. You can integrate MCPS and similar stuff as well. This is also quite scriptable and you can use it in your workflows which is pretty good. Now that's majorly it now let's check it out.
So first of all you'll have to install it with this command. You need to have Python and UV installed. You can refer to the links on the GitHub repo for how to install them. It's simple. Then you just have to run this command and you'll be good to go. It will install automatically.
But before we do that, let me tell you about today's sponsor, Photogenius AI. Photogenius AI is an all-in-one AI powered creation suite that lets you type anything and get stunning visuals instantly. Now also the best place to use Google's Nano Banana for images and V3 for videos, plus affordable 3D model generation. Inside the image playground, Nanobanana shines for fast, highquality image generation. and you can add reference images and do edits right in the tool. You also get Flux, stable diffusion, Kandinsky, and more in one place. The video playground supports Google VO3 with and without reference images, and you can render in different styles without the usual complexity. Great for coders who want results, not knobs. For 3D, you can upload a PNG, think a Lego build or a simple robot, and get a printable model. Cheap, quick, and surprisingly clean for rapid prototyping. Pricing is among the best for V3 and Nano Banana. And you still have access to about 10 other handy AI tools like avatars, background removal, logo, emoji, ads, and app icons in the creative tool suite. It starts at a low entry price, and you can take an additional 30% off with my coupon code king30. Check Photo Genius out through the link in the description and try it for yourself.
Now, back to the video. To start it, you'll just need to run the Kimmy command and it will launch. To be honest, it's very functional and I like the whole aesthetic that it has. When you start it, you'll probably have to configure it. To configure, run the / setup command here, and this will give you the option to choose whether you want to use the Kimmy coding plan or the Moonshot general API. If you use the coding plan, go to the settings of your KI account. Then head over to subscription. Here you'll see the KI for coding option with your limit. You can click this and it will open up the API key that you can use. Just paste it in and it should work fine.
Now the first option is that if you press controll + k it will toggle between the shell and agent mode. In shell mode it will show you the dollar sign and whatever you type here will be executed just like in any other command shell and it works pretty well. In agent mode it will show all its AI related stuff. It's kind of cool and I can see people actually using this as their main shell. I won't do that though, at least not yet.
Anyway, there are some more slash commands. There's the clear command to clear the terminal output. There's also the compact command to summarize the context up until that point and free up your context window. You also get the debug command, which is something new. This basically shows you all the logs of the messages and context and allows you to see what context it has exactly. So, this is cool. Another one that I liked is the release notes command. After each update, you can use it to quickly see the new update changes. That's majorly it.
Now, you can just send a message and the tool calling and other features work pretty well with this. You get the same options for tool call approval and things like that. Kimmy is quite good at tool calling, so this works really well. Though the model capabilities are a bit limited, at least in the tasks that I tested it on. I've heard a lot of good things about how it can be great at planning bigger tasks, doing awesome debugging, and writing some in-depth code, and it makes sense. Kimmy has that typical big model power since it's a one trillion parameter model and everybody knows that Kimmy is just awesome at writing as well. I've used it a lot for planning and coding too. You can get this for $20, which makes it a good deal for Kimmy lovers for sure. You also get access to features like deep research, agent mode, and other great tools.
Previously, I used to say that you could either use GLM for most tasks since it's much cheaper or go for Synthetics $20 plan. That's also a really good plan. It includes most of the open models and you get quite a good limit compared to Kim's own plan. However, Kim's official one might arguably be better for tool calling, though synthetic is also really great. So, if you're just looking for higher limits, then synthetic might also be a good option. But Kimy's interface and web tools are in themselves worth it for $20. And the model is really good. So, I can surely recommend you to check it out and use whichever version of Ki you like. Overall, it's pretty cool.
Anyway, share your thoughts below and subscribe to the channel. You can also donate via super thanks option or join the channel as well and get some perks. I'll see you in the next video. Bye.