Technical Context
I'm not focusing on another debate about note-taking apps, but on a tangible engineering gap in the market. When I build AI automation on top of an internal knowledge base, I need more than a pretty graph and markdown. I need backups, a transparent data format, and encryption that isn't just "wrap the folder in VeraCrypt."
I looked at the usual suspects. Obsidian is convenient because it stores everything in .md, which is easy to back up, run through git, and migrate anywhere. But there's no built-in encryption, meaning the security model is cobbled together from external workarounds and plugins, and that's where I start to cringe.
Anytype comes from the other direction: encryption is there, the local-first approach is solid, and the idea is appealing. But its data storage format is proprietary, which immediately breaks some of my use cases that require a proper graph knowledge base for projects without being tied to a specific product's internal magic.
Logseq and TiddlyWiki also don't fully solve the problem. The first is great for developers, especially if you like a query-based approach and plaintext, but it has the same encryption story: wrap it with external tools. The second can be encrypted, but it's a compromise on scale and convenience, especially if the knowledge base starts to act as infrastructure rather than a personal archive.
And here's the most interesting part: we have graph databases like Neo4j or Memgraph, but that's a different level of abstraction. They don't provide a ready-made PKM layer for someone who wants to think in terms of knowledge, not set up a separate interface, sync, editor, search, and access rights.
Impact on Business and Automation
For a business, this isn't an academic trifle. If your knowledge about projects, clients, R&D, and internal processes resides in a system without a proper data ownership model, AI implementation quickly becomes fragile.
Teams that need control over a private knowledge base and proper AI integration without vendor lock-in are losing out. The winners will be those who either assemble a neat open stack around plaintext and external encryption or are the first to release a sensible product at the intersection of PKM, graph, and security.
At Nahornyi AI Lab, we regularly face these architectural crossroads with clients: what to store in markdown, what to move to a graph, where to encrypt, how to handle backups without sacrificing usability. If your knowledge is already accumulating faster than you can protect and use it, let's break it down into layers and build an AI solutions architecture without unnecessary magic and without the risk of losing your most valuable asset.