I don’t really understand exactly what you’re doing here, but I am really interested to find out.
So is this a bit similar to how we are having LLM memories, where they remember information from past conversations (RAG)?
I think they were some dating AI bots that tried to do this more intentionally. They would have a section called memories, and that would be a list of the memories they had formed that could be edited.
In SillyTavern, this is done using a law book, but it is a bit troublesome to edit and I think the conversational chat interface you have here using Discord looks a lot more usable.
I really like how it prompt you for more information.
I specialise more in using a Zettelkasten for bottom-up thinking, but this looks like a way to interact with your note that potentially Zettel notes per memory. The only difference is that rather than just saving a memory, we want to see how this idea could be transformed into a permanent note and connected to other notes.
Hey — appreciate this a lot. You’re totally tracking in the right direction.
Yeah, the memory structure here is inspired by approaches like RAG, but it’s aiming for something a bit more reflective. Less just “remember past context,” more “curate meaningful memory artifacts.” Think: not just what was said, but why it mattered, who was involved, what it connects to. Sort of like structured Zettels — but formed through conversation.
The dating bots and SillyTavern-style memory modules definitely cracked open the door. What we’re doing differently is treating memory as a first-class object: it’s parsed, titled, tagged, and threaded in real time via agents. And yeah — the prompt-back design is intentional. It asks, not just logs. That makes the memory participatory, not passive.
Your Zettelkasten connection is on point. In fact, one of the directions we’re exploring next is memory stitching — so over time, scattered moments can be linked into coherent arcs. Not just storing notes, but showing how they evolve into permanent knowledge.
Would love to hear more about how you use Zettelkasten. Maybe there’s a crossover path here — agent-supported note evolution?
Thanks again for digging in. This kind of thoughtful curiosity is exactly why we’re building in public.
There is '5W1H' which seems like your current memory framework, but going deeper into how you feel, why, or what happened before this, what you think will happen next, is this likely to happen again, why this is a systemic problem etc could be interesting.
I like obsidian backlinks and graphs to see how ideas are linked. The Strange New Worlds plugin, and the smart connections plugin allow you to see how notes are linked to each other right now and potential other connections based off similarity of the notes. That might be something to add to the memory system, a suggestion of similar events. Inbuilt discoverability & connection.
This is fascinating.
I don’t really understand exactly what you’re doing here, but I am really interested to find out.
So is this a bit similar to how we are having LLM memories, where they remember information from past conversations (RAG)?
I think they were some dating AI bots that tried to do this more intentionally. They would have a section called memories, and that would be a list of the memories they had formed that could be edited.
In SillyTavern, this is done using a law book, but it is a bit troublesome to edit and I think the conversational chat interface you have here using Discord looks a lot more usable.
I really like how it prompt you for more information.
I specialise more in using a Zettelkasten for bottom-up thinking, but this looks like a way to interact with your note that potentially Zettel notes per memory. The only difference is that rather than just saving a memory, we want to see how this idea could be transformed into a permanent note and connected to other notes.
Hey — appreciate this a lot. You’re totally tracking in the right direction.
Yeah, the memory structure here is inspired by approaches like RAG, but it’s aiming for something a bit more reflective. Less just “remember past context,” more “curate meaningful memory artifacts.” Think: not just what was said, but why it mattered, who was involved, what it connects to. Sort of like structured Zettels — but formed through conversation.
The dating bots and SillyTavern-style memory modules definitely cracked open the door. What we’re doing differently is treating memory as a first-class object: it’s parsed, titled, tagged, and threaded in real time via agents. And yeah — the prompt-back design is intentional. It asks, not just logs. That makes the memory participatory, not passive.
Your Zettelkasten connection is on point. In fact, one of the directions we’re exploring next is memory stitching — so over time, scattered moments can be linked into coherent arcs. Not just storing notes, but showing how they evolve into permanent knowledge.
Would love to hear more about how you use Zettelkasten. Maybe there’s a crossover path here — agent-supported note evolution?
Thanks again for digging in. This kind of thoughtful curiosity is exactly why we’re building in public.
— Carl /build🧱
Thanks for the extensive reply.
Here is a quick guide to how I use the Zettelkasten: https://creatorpreneurteststudio.substack.com/p/7-minute-zettelkasten-kickstart - You can see there, I tend to use my Zettelkastenmore intuitively rather than sticking to any specific ideating framework. I think the Idea Compass (https://www.youtube.com/watch?v=5O46Rqh5zHE&t=284s by Vicky Zhao and Fei-Ling Tseng) or Dan Koe's Korenotes framework also give good direction for prompting people.
There is '5W1H' which seems like your current memory framework, but going deeper into how you feel, why, or what happened before this, what you think will happen next, is this likely to happen again, why this is a systemic problem etc could be interesting.
I like obsidian backlinks and graphs to see how ideas are linked. The Strange New Worlds plugin, and the smart connections plugin allow you to see how notes are linked to each other right now and potential other connections based off similarity of the notes. That might be something to add to the memory system, a suggestion of similar events. Inbuilt discoverability & connection.