Some of these suggestions are related to the interface itself (ACC), not the model (AI) that runs on the server. Modding ACC's features are possible since the code is open for anyone to edit. The rest will be only significantly improved when the model is upgraded. In my view, ACC is more of a generic platform to chat with the bots in several separate threads, not really one that is focused solely on being an RPG adventure (like AI RPG for example). Something like this might be a different goal than what ACC intends to be. An upgrade to AI RPG with these features would sound more appropriate, I believe. That is why AI Dungeon is called that, it is focused on a specific type of AI roleplay and usage. ACC is a general chatting platform, and more advanced users have tools available to modify it to their use case, similar to SillyTavern, for example. Most of the AI problems you mentioned have to do with the low context window (6k tokens) it has. It's not able to ingest that much information, so unless it is properly curated with what is happening, the AI will miss things and details. Many other models on the market have immense context windows, from 64k or 128k to 1 million (or so Google claims). Until the text model is upgraded, the best you can do is curate the information available to it (past messages, summaries, character descriptions and General Writing Instructions are the most important). Now, about the AI issues, let's start with chars not leaving the scene. I faced this problem a lot more with narrator characters or using a single character. One thing to keep in mind is that the description of the main AI of the thread is always sent to the AI on every message. So, if you're having an actual character as the main AI, even if they left the scene, their description is always sent. This can cause the AI to keep referring to them, especially if their description is long. I started using a narrator character (more of a "hub" character just to host common lorebooks and other things) and never posted as them, but with my main chars. This description was always sent to the AI and improved the narrative instead, but on very rare occasions it would still refer to them as "The Narrator". Using multiple chars made this a lot more consistent, and the AI was able to read previous messages to better decide what was their status. Had a long 2v2 battle between chars, and once two of them died, they stayed dead, had no issues there. Make sure to always track your summaries and memories to avoid wrong information being fed to the AI. I eventually found out multiple chars in the same thread causes all their descriptions to be sent (for the chars that posted on the last 20 messages), so this made the AI receive more information about them instead of it being only collected from previous messages or only occasionally retrieved if on the lorebook (which has a low limit of how many entries can be retrieved). Perchance uses an old text model, curating the information, understanding its limitations, and knowing what information it's receiving helps improving you dealing with it better and improving your experience. You'll have to micromanage? Sure, but the AI is not magical, it has limitations that can become apparent after a while unless you work around them. There are several tricks available like using character injection or hidden system messages. Eventually, I decided to code a script to adjust which char description was sent, so that I could conserve tokens and also better indicate that a character has left the immediate area. This solved the problem of chars being mistakenly referred to as if they were still around. It is available on my fork that VioneT posted, at the options bar on the bottom right.