A portrait, not a transcript

Every AI assistant either forgets you between sessions or keeps a chat log. Neither of those is knowing you.

We read dozens of essays on self-organisation and took apart two dozen competitors. For almost all of them "memory" is either nothing, or a RAG search over your own notes that rebuilds every few minutes. The closest in spirit is Morgen with its Frames — but Frames are planning templates, not an understanding of the person.

What we accumulate

Understanding at three levels — you, your spheres, your projects. "Underestimates time." "More productive in the morning." "This project is stuck on an external dependency, not on you." These are digested, typed, versioned facts — not a dialogue log. You can open, read and correct them.

The portrait changes behaviour, not just tone

New in 0.7: planning facts. "Do not plan in the evening", "Monday — lower capacity". They come from watching your rhythm; the system offers to write such a fact down — it does not push, it waits until you open the right place yourself — and you confirm. From then on the fact is read deterministically when the day is assembled.

That is adaptivity. But our way: a readable, editable fact, not a hidden weight adjustment inside a black box. We do not switch methodologies — one system for everyone, what adapts is the contents of the memory.