The Doxwell Story

Doxwell is my alter ego — my antipode.

I tend to make chaos in my documents. I run several lines of activity in parallel, and the German postal system keeps a steady stream of paper mail flowing in, all of which has to be read, understood, and answered, often quickly. The chaos piled up.

I'm a software engineer by background, and earlier in my career I'd spent several years in a document processing group — so I wasn't a complete beginner at the problem. I went after it with software, and from the start I had a specific picture of what I wanted: a system with a built-in time machine, real version control, entity relationships that hold across documents, and access control with roles. Not the things I saw in existing tools.

But a system that all-inclusive is big, and big systems are resource-demanding. The resource mine demanded most was my time — the same scarce thing that had caused the pile-up in the first place. So for two years I made small steps in the slivers my primary affairs left over: mostly scans and OCR. Then this January I gave it a boost, helped along by what people now call vibe coding, and it inspired me to complete the task.

I read a lot during this stretch. I'd never stopped programming, but I hadn't had much time for it either — and this field moves so fast that even a moderate pace puts you behind. I had ideas to test from years of collecting them. Some held up. Some didn't.

Of course I had doubts. The industry might already solve this better than I do. So part of my weekly rhythm was looking for articles and asking my AI assistant to compare what I was building against what was out there. I was actually proud — the system was outperforming everything I knew. But the field is moving fast, and I don't have much time before the industry catches up.

I started talking to people. Some pointed me to work I hadn't seen. It turned out I'd independently arrived at things close to ideas in HippoRAG, the time-aware structures of certain temporal databases, and even the spirit of Karpathy's now-famous LLM Wiki — but at scale. Some of those parallels survived as core parts of the system. Others didn't.

After many successful uses of the system — moments where it just made my own life much easier — I realized this is probably a very common problem to address.

That's why Doxwell exists.