Ever since I started using computers, my digital footprint has kept growing:
- Sources in electronic form: PDFs, ePUBs, links, my own DOCX files, and so on.
- My own artifacts — texts, presentations, and the like — with carefully developed content and, where possible, references back to the sources.
- More recently, AI chat sessions that have led, through dialogue, to surprisingly robust insights.
What I never had:
- The ability to search across all my sources for relevance — or at least proximity — to whatever question I was working on.
- Even when I did find a match, a way to mark the relevant passages so they could actually be referenced later.
- A system that could help surface cross-connections between materials.
Just about every PKM tool I tried promised exactly that — and I tried most of the major ones over the years: Obsidian, Notion, Scrintal, and others. I had eventually settled on Scrintal, which I genuinely liked aesthetically. Then it died. All the data I had entered there is gone — maybe it’ll resurface someday. But the experience left me with a firm resolution: never again get locked into a vendor. My data belongs to me, lives on my own storage, and exists as plain text files — Markdown at most. As it turns out, the current state of technology makes this not just possible but practical.
Let me briefly sketch what my current setup looks like — details will follow. Here’s what it can do:
- All data is stored locally or synced to a cloud service in a plain text format. PDFs are fine too — they’ve essentially become a standard at this point. Nextcloud is particularly worth recommending here: it’s independent of any platform’s whims, and you can host it wherever you like (mine runs on German servers).
- The intellectual heavy lifting is AI-assisted. I’ve come to rely heavily on Claude for this — with one small caveat: it’s not based in Europe.
- The technical side is largely automated:
- Building a coherent logical file structure
- Establishing cross-links between related materials
- Maintaining an overview and enabling search across all references
After about one to two days of migration work, the whole thing runs beautifully — and it’s future-proof. The plain text files can be read and processed by any AI system down the line. Obsidian, which I now use almost exclusively as a frontend, could be replaced by something else if needed — some people prefer Visual Studio Code. Obsidian is based in Canada, for what it’s worth; everyone can decide what they’re comfortable with.
The bottom line: the system now works better than anything I ever managed by hand. After migrating a test project, I’ve decided to start all new material here directly — and to migrate older content as soon as it’s needed again.
Concrete how-tos will follow. For now, consider this a first glimpse of where things stand.
