This is whats coming Guys

[CodeAbra]

[Marsu6996]

This looks genuinely cool, Areg. The visual direction really makes the project feel alive, and I can see why this would make iai-pme much easier to understand for normal users.

My practical user point still stands, though: the feature I would value most is being able to understand why a memory surfaced. Which memory, what nearby links or clusters led to it, and whether it came from episodic, semantic, procedural, or fading memory.

So for me the visual map is the exciting part, and the practical win would be trust and debugging: daemon awake/asleep, last consolidation, memory counts, and a simple way to inspect or correct a bad memory without digging through the DB.

Happy to test this on my setup when you have something rough. I use iai-pme daily through MCP, CLI, and Claude Desktop, so it should be a decent stress test.

[CodeAbra]

You will be able to put any memories you want in a queue to fade

[danielhertz1999-bit]

Oh hell yeah! That looks absolutely insane! I really dig it

[CodeAbra]

Just updated
Please don't hold back on criticism, issues and PR's!

Hope you guys enjoy this version even more

Ask me any questions

Im still in the process of writing the new Readme with instructions so some stuff might need explanation

[CodeAbra]

Let me know your thoughts about the app!
Would love to get some feedback
I don't really think I like the ux/ui experience but im not sure what im missing here

[Marsu6996]

Hey Areg 🙂 first off — this is lovely, congrats.

I genuinely love the idea, and I mean it: I keep it open full-size on a window right next to my work stuff, just for the pleasure of watching it come alive. It's beautiful, and it makes the whole memory concept feel real and tangible in a way nothing else does.

If it helps at all with the UX question, here's my honest little bit — and please take it gently, because the visuals themselves are great. I realize I look at it far more than I use it. I don't often have a concrete reason to actually reach into it, and I think that's less about your design and more about the nature of a memory engine itself: it's built to run quietly in the background, so there isn't naturally much to do in the foreground.

So maybe the thing that's "missing" isn't visual at all — it might just be a few clear reasons to open it on purpose: curating memories (pin the important ones, send others to fade — like you mentioned), or peeking at why a memory surfaced (which cluster/links pulled it up). Give me one or two moments where opening the app is genuinely the best way to get something done, and all this polish you've already built will really shine.

Either way — truly nice work, and thank you for sharing it. Happy to brainstorm on this anytime if useful 🙌

[CodeAbra]

I think you can do all of those things in the app. Try zooming and clicking on the memory

[joerybka]

Hey, this is pretty cool, but I did have some trouble navigating a bit trying to find nodes that were interactable. I have attached a video, sorry about the tearing effects, I was using ffmpeg directly to screen record and it was less pretty than I was anticipating.

Anyway, the first part of the video is me trying to find nodes to interact with, it was not immediately clear to me what was a clickable node vs what was background. You can see me hunting around in the cloud for something to click on, largely watching the hand turn into a pointer and then trying to find what it did that with. Right near the end of that, I watched one of the pong ball's trajectory and where it terminated and was able to find that node to click on. I am not sure if there is a way to better highlight or differentiate nodes that are interactable.

The second part is that I was not finding an immediate way to zoom in, pull the brain to a specific section, and keep it there. The brain kept elastic pulling back to the middle such that it was not readily apparent how I could zoom in on the periphery of the brain instead of the center.

brainInteraction_compressed.mp4

Some smaller things:

• The search in the upper left, it would be nice if it had an "X" to the right end of the input box so that it was easy to clear
• When I select a node and then grab the screen to view the connections with other nodes in 3d, and then let go, the selected node is deselected

Things I like:

• There is a recenter function, you would be shocked the number of people that forget that
• the colors are generally pleasing and differentiated, at least to me having normal color vision

One thing i was unsure about: the "Working for you" section was all 0s, not sure if i just dont have enough data yet for that to be functioning or if there is something wrong with the display.