Personal Platform

This is the second version of a personal platform I'm building for myself. I started an earlier version, and instead of patching it forward I stopped and rethought the approach. The earlier version's actual failure was that everything was too tightly coupled, when one of the core ideas is being able to work on features independently and add or remove them easily. Memory ownership was split across two languages, so the boundary between what each side owned kept leaking in practice. There was also a single do-everything call for fetching context that ended up needing a type switch to handle every caller, and a hand-tuned relevance formula with fixed weights I had no way to validate. None of that carries over as code. A few of the underlying ideas do, but rebuilt properly.

The idea is that the platform lives and grows with me for years, instead of solving one problem and getting abandoned. Whenever I come across something worth adding to how I work, whether that's a new tool, an automation, something embedded, or a piece of software, this is where I plug it in. Each one becomes its own isolated feature with its own frontend and backend, connected to the rest only through a couple of narrow contracts. What makes it a platform and not a pile of unrelated apps is a shared "brain": a memory service that features can send events to, and that uses AI offline to build up a structured understanding of me over time, instead of each feature keeping its own guess.

So far this is all architecture and design work. I've worked out how the core, the brain, and the features should be isolated from each other, and how authentication and multi-user isolation should work even though it's just me today. I also have a long list of feature ideas the design needs to hold up against, without committing to any of them yet: projects, learning, a wiki, reading and books, a speed reader, and controllers for other devices, among others. None of it is built yet, and the boundary between core and features described in the docs is a plan, not implemented code.

The next step is the first milestone: a small core I can finish (accounts, auth, routing), built in parallel with a small brain (profile, events, basic retrieval). The two are connected only by a minimal placeholder feature, whose whole job is to prove that the contract between them works.