I personally use Copilot and Gemini. The first one can be used if you go to Bing from MS and then choose the AI there, then you have the free Copilot available. It is able to remember VERY long chats over weeks to months, as long as there is not too much time between two sessions, but even then it remembers even the last things of the discussion. You only need to keep the tab open, but you can close the browser (as long as the option to start with the same open tabs is enabled). The tab contains the session info. If you do not log in with a MS account, it will ask from time to time to login and to verify if you are a human (clicking a checkbox), that's all. With a MS account you can save multiple chats and return to each of it later. Similar with Gemini, which can be started using Google search with KI mode. Both AIs are similar in what they can do for you. Gemini has, as long as you are logged in with a Google account, a new AI named "Coding Assistant" which is better in coding than the "normal" AIs. In VSCode it makes not really a big difference if you use the chat options. Because copy/paste can be done from a browser window also and doesn't use space in the VS Code window. But: You can enable the Copilot in VSCode for coding help directly in the code window, that DOES make a huge difference! I have tested it with developing addons for Blender and while you're writing code in the code window, the AI in the background analyzes the code and recommends i.e. comments, code and even complete code blocks for what it thinks may be the right direction following the contents of the existing code/comments. This is REALLY surprising in many cases, you'll often ask yourself, what is it what I already wrote what let the AI think this could be the correct completion? It is limited in the number of free proposals per month, but the limit is big enough for most cases. And of course: Gemini and Copilot have a huge server farm behind so they are really fast in execution. Other models are often longer thinking about a problem. I've already tried models like Qwen, especially locally installed, but they have not the same professional results as the two big ones. Especially remembering a longer discussion they are not good. Something you really need when you develop a bigger codebase like an addon (more than 3 cubes in a stack... ;) ).

Sure, but indeed I personally don't let any agent on my computer if it is not, like the Copilot VSCode one, already implemented into the product itself and which has no other possibilities than to help coding. I thought you've said that you copy/paste the proposals from the chat window in the sidebar. Maybe I should have watched the video to the end.. :D Yes, right, it's slower, but also more safe. I've tried let Gemini changing the code directly (it has offered a code window in the browser where you can paste your code in and it can directly change the code there). It was really bad and it is also in chats, as the AI very often cannot follow the thoughts exactly and create whatever it thinks could be good. You need a lot of explanations to the AI until you get at least a nearly good result in case of more complicate codes. You can even send it the complete code file and it tells you that it has read it and understand it - and it doesn't. Because the AI is not a human it only tries to get the nearest result, best as possible, nothing else. The longer the code is the less it understands, so best is to work with it together with smaller functions and explain only what this one should do, tell it a bit about the complete target and the results are very much better. It can concentrate on the more short functions a lot better. I worked with the AIs in Blender/Python now for more than 1.5 years, very deep, last year nearly the whole year at least 8 hours a day. The AIs got a lot better in that time and also what they can do as they do not only learn more and more, the developers also improves their functions and way of working every day, so the results are nowadays really a LOT more impressive than they were a year ago. But at the current time from my experience I can say that I would not let the AIs do the whole job alone, the results are frustrating - as I said, with more complicate addons. For simple jobs they can of course create a complete addon because there is enough input to learn from in tutorials which they can use.

Hi! Sure, usually you can access each and every information from the Blender API, although often not easy to achieve. I have no experience in case of bones, but other's have... :) https://blender.stackexchange.com/questions/111314/how-do-you-access-and-set-a-specific-bone-animation-fcurve-in-python-e-g-set Please keep in mind that this is an older entry, Blender's fcurve API has changed in the meantime. The newer one is not easier, but needs more investigation to understand it. You should look at the official API documentation and asking an AI is also a good way.

There are already addons to manage n-menu entries. Some samples: https://superhivemarket.com/products/cool-addons-menu https://superhivemarket.com/products/simple-n-panel Moreover, the layout possibilities Blender offers for the self-made UI are really immense. You can layout whatever you want in columns and rows you can define by yourself, you can also use sub-elements like a list where you can organize columns on many ways (including graphics) and you can also use graphics to be displayed in an n-menu which can be programmatically created or loaded from the file system. As you can use anythiing Blender supports this includes to render a gray image of your character, save it as picture, load it into the UI as icon - lot of programming work of course, you need to learn a lot about how to do all that. You can render all the buttons in the sample in similar way, but the heavy part would be to render them on exact positions over the image. Needs a lot of experiments (and frustrations... :) ). But surely you could get something similar. Expect some months of programming/experimenting... ;)

...und nicht zu vergessen: Frohe Weihnachten und guten Rutsch!... ;)

I recommend to start with something like this page: https://www.w3schools.com/python/ Step by step with easy steps and you can try it out by yourself directly on the page.