MyMemory · under the hood
How Darwin remembers
Whatever goes into Darwin's permanent memory stays as plain text notes on your computer. Before every answer, Darwin automatically picks the parts related to the question. Nothing gets retrained or fine-tuned — Darwin looks things up and simply remembers.
One memory, two forms
The vault: notes in Markdown you can open and edit at any time.
The index: a map of meanings built from the vault, which can be rebuilt from scratch whenever needed.
The index is derived. The vault stays the source of truth.
Schema 1 — One truth, one map
What happens when you ask something
Darwin first automatically picks a few of the most closely related pieces of memory. When that isn't enough, he searches deliberately — by meaning or by exact words. If he needs wider context, he opens the whole note.
So a single answer never carries the whole memory — only a small selection of what's relevant.
Schema 2 — The path of one question
Why this is different
The source notes, their embeddings and the meaning index all stay on your computer. Turning text into numbers is done by a local model.
The question itself and the selected context are then processed by the AI model of your choice. If the whole flow has to be fully offline, use the local model for that too.
Darwin recalls automatically, not on a special command. His critical rules are always with him. And you can read, correct, back up or delete the memory — it's ordinary files.
Schema 3 — Why local
pieces of your documents travel to the provider
only a thin arrow leads out: the answer itself
He remembers your conversations too
Besides notes, Darwin can (optionally) keep the conversations themselves. Ask a month later "how did we solve that back then?" and he finds it — even though nobody wrote it down.
Conversation snippets are treated as a less trustworthy source than notes. A conversation also contains things that later turned out to be wrong, and Darwin must not recycle his own old guess as a fact. What should hold permanently belongs in a note — a conversation is a record, not a conclusion.
When he gets something wrong, you fix it in a minute
The memory is plain text files. If Darwin memorises nonsense, you rewrite one sentence — and within five minutes it holds. No model retraining, no waiting, no service call.
That's exactly what assistants living in someone else's cloud usually can't offer: once something lands in their memory, the user never sees it, let alone corrects it.
What it looks like in practice
"What did we agree with the client in March?"
Darwin finds the note even if the word agree never appears in it — the search runs on meaning, not on matching letters.
"Remember that invoices always go out by the fifteenth."
One sentence is stored and holds immediately. Five minutes later Darwin builds on it in a completely different topic.
The numbers
Schema 4 — Why the archive won't bury the answer
What you get out of it
You don't have to remember what you've told Darwin and when — he returns to stored information automatically. Local memory search costs you nothing extra. The source notes and their index stay with you.
The memory is yours: open it, back it up, correct it, or take it with you.
See the memory with your own eyes
Every note is a node in a living graph, and meanings form a map you can search — right on your machine.
See the living memory map › Get Darwin — from €199