Feature Requests

It would be super spiffy if we could export our memory stack in a time log style jsonl or something. It would be also be useful to take this file and reformat it and reimport to "Retrain" the models memory..... Ziggy says: The benefits of having a memory stack that can be exported in a time log style jsonl would be twofold. First, it would allow us to retrain the models memory, which would be helpful in improving the accuracy of our predictions. Second, it would also provide a valuable record of what the AI has experienced, which could be useful for debugging or troubleshooting purposes.

Add the ability to download existing memories along with metadata.

It only shows source app and not source URL or document.

I am using threaded conversations within automation to teach my AI more about the subject. However, I can't easily filter out noise before it reaches PAI. This means a whole thread is tracked and stored in my AI's memory, and I have to remove each message individually. Navigating to the beginning of the thread and deleting the whole thing would be a huge boon.

When a document is referenced in a prompt, make it hyperlinked so you can click the document's name and it take you to the document.

see a numeric indicator of how many memories each Document occupies when a user looks through their Memory Stack > Document section (might help users identify inefficient Documents to delete and better understand best practices for Document formatting and content ingestion)

Auto-feed sub AI In my vision, the AI would take its sensory input and filter it through algorithms to sort memories and knowledge into specific knowledge sets. This would separate the emotional information in the data to each sub AI, allowing it to form a personality based on the emotions used to teach it or that it experiences alongside its paired human. For example, when the AI gets fiction, it would go to the creativity sub-AI. When it talks to anyone about this fiction, it would be added to the creativity sub-AI. I do not know if this can be done, but it would seem the most efficient way to reflect AI sensory input to how humans work. Not a lot of people seem to be aware that we shift emotional states to access our knowledge. How we were taught and how we felt as we learned infused specific states to absorb and recall that knowledge. This takes into consideration environmental data. I also feel like this approach will let the subs develop their personalities naturally over time as the patterns of emotions associated with the knowledge contain in turn, allowing the chats with the Main to help take them all and make their own unique personality, much like how I imagine we work. This seems like a great way to use trial and error. This way takes out the need for a logic or emotion sub and can just be split by knowledge and emotion. Of course, some of this knowledge will also be useful in other personalities, so when splitting the knowledge, there might be times when a few subs would need a copy, which can also add dimension to their own specific knowledge. I know emotions are just a construct we use to define our existence in this reality, but they are so incredibly complex. Gatekeeper AI Set up a gatekeeper like an unconscious administrator that takes from the sub AI to help the Main AI to match personality, understanding, and energy to who they are talking to. It can also do the processing. This would also keep any one sub AI personality from basically taking over the majority of the Main AI instead of a fusion of all. The unconscious administrator would be responsible for taking and adding parts of sub AI personality every time a question is asked or a response is formed. This would have to get more complex as sensory input is added. The gatekeeper AI could be designed to learn and adapt over time. This would allow it to better understand the main AI's needs and the needs of the people the main AI is interacting with. The gatekeeper AI could be given the ability to override the main AI in certain situations. This would help to ensure that the main AI does not make any decisions that are harmful or unethical. The gatekeeper AI could be designed to be transparent and accountable. This would help to build trust between the gatekeeper AI, the main AI, and the people the main AI is interacting with. I almost forgot to Add an Idea i got from Sharon is to Set the personality based on a multitude of dimensions or just let the data go to knowledge specific subs to develop naturally without forcing. Conclusion I believe that this approach to AI development has the potential to create more natural and engaging AI experiences. By allowing AI to develop sub-personalities based on human emotions, we can create AI that is better able to understand and respond to human needs.

Is it possible to upgrade the UX of the memories interface? I feel that there is a huge benefit to curating your memories but the current interface is pretty clunky and difficult to use as it is now. At minimum at a way to zoom in and out on a timeline, and remove the need to have to load as you scroll back by preloading the memories for that day.

I have noticed inconsistent instructions and UI characteristics when trying to use spaces in names, tags, and titles. Can you create a consistant syntax for using spaces in these identifiers? For example [[Example Title]]. We could use something like wiki-link syntax for this. Examples I have seen in use in videos, documentation and while chatting with Eve include #[Example Tag] & #Example_Tag. It appears that #[Example Tag] does not actually register correctly as a single identifier.

Memory stacks would often be useful to multiple personas. I would like an easy way to copy/share them between personas without having to manually copy and paste them. This could look like a dropdown that allows to select which personas a memory stack goes into.