Smart Second Brain: six design questions worth asking

An amazing initiative to build a smart second brain, which is now in its second iteration on GitHub. It makes me wonder about a few design choices compared to how the biological brain actually works.

• Search: I can see that the plugin's search uses # and / to look up tags and folders. Who manages those — creating new tags, deciding on folder structure, keeping it organised? That's manual overhead that doesn't go away.
• Folders: One structural problem with folders is that each note can only belong to one at a time. That locks you into an information architecture you can't fully predict in advance — your relationship with a PKM tends to evolve in ways you don't anticipate early on.
• Smart graph: Is the graph primarily useful for seeing which topics have more notes? The standard Obsidian graph doesn't feel scalable. Once you reach thousands — let alone tens of thousands — of notes, it becomes difficult to read any meaningful pattern from all that noise. It feels like we might need some sort of index of the most popular collections with summaries per collection.
• Agents: I understand the meaning of skills, but could you expand on how you realise memory specifically? That sounds like a significant feature, which potentially solves a lot of problems with the context window. And separately, when you mention reading, editing, and deleting memories, those sound like manual operations — which raises the same scalability question as the graph.
• Chat files: Why introduce a new .chat format rather than using conventional .md? There's a growing view that .md may eventually give way to .html — how do you benefit from introducing a new file type versus following a convention?
• Agent-agnostic: When you mention agents running on different models, does that mean the system is effectively model-agnostic — that the underlying intelligence is swappable? That sounds super exciting, meaning that potentially we could have agents working independently within a specific PKMS.

Would be genuinely curious to see the devs address any of these — especially memory at scale and what the graph looks like past 10,000 notes.

Until the next sync,
Zhenia