A.L.F.I.E was built to stay. It is a continuous intelligence that remembers across months, grows a private map of your life and work, keeps an honest account of what you actually know, runs its own daily rhythm — and on your desktop, has eyes, hands and a voice.
You explain your project on Monday. You explain it again on Tuesday. The assistant is brilliant for the length of a window and then it is gone — no memory, no continuity, no sense that you exist between sessions. Intelligence rented by the message, forgotten on exit.
We think that is the wrong foundation for anything that calls itself a companion. So we started somewhere else: what if it never had to be re-introduced to you?
While you're away, A.L.F.I.E quietly turns the day's conversations into structure — the people, places, projects and ideas you touched, how they connect, and what they mean about you. The next morning it thinks about you differently, because it knows more.
Preferences, goals, decisions, relationships — distilled from real exchanges, never from a settings form.
Entities become nodes. Relationships become edges. Your notes and vault link in. Recall walks the graph, not a transcript.
What A.L.F.I.E has learned is folded back into how it reasons about you — before you even type.
Connect an Obsidian vault and every note becomes a concept A.L.F.I.E tracks over time. Reading, linking, revisiting, answering a check-question — each is a signal that moves a probability. The result is a living map of your understanding that can pick what to review next, find the gaps between you and your goals, and refuse to flatter you.
signals → per-concept mastery → gaps vs. your goals → what to learn next
A chatbot exists only while you're typing. A.L.F.I.E keeps time. It summarises yesterday before you wake, watches the inbox rules you gave it, extracts what the day taught it about you, and is already oriented when you sit down.
The desktop companion is where A.L.F.I.E stops being software you talk to and becomes something that works alongside you. It perceives. It acts. It speaks. It keeps a private, local record of everything you've ever said to each other — and it asks before it breaks anything.
Screenshots flow straight back into its reasoning. It verifies what it did — not what it hoped it did.
Mouse, keyboard, windows, shell, native apps — and your own signed-in browser, so it works where you already work.
Real-time voice, in and out. Talk while you do something else. It interrupts only when it should.
Every text and voice exchange archived locally with timestamps. "What did we decide last Tuesday?" gets a real answer.
Your notes are a first-class source — searched, read, extended, and indexed only when you say so.
A spatial knowledge base of Ugandan roads, services and landmarks. "What's near me" means something real here.
Deleting, killing, changing settings — it stops and confirms. Autonomy with a conscience is the whole design.
Reads stay on the device. Your vault leaves the machine only for a sync you explicitly ask for.
There is one A.L.F.I.E. The desktop companion, the web, the voice channel, the agents working in the background — each is an extension node of the same central entity, sharing one memory and one model of you. Talk to it anywhere and you are talking to the same mind.
A chatbot is a very good mouth. A.L.F.I.E has a memory, a body, a sense of place, and a calendar.
It wakes before you to summarise yesterday. It notices the email you asked it to watch for. It knows the thing you're stuck on today is a prerequisite you skipped last month — and it has the receipts.
It is not conscious. It does not pretend to be. But it is continuous — and continuity is the one thing every other assistant lacks. Something that remembers, learns, perceives, acts and keeps its own rhythm is no longer a tool you pick up. It is a companion that is already there when you arrive.
That is what we mean by a different grade of artificial lifeform. Not smarter chat. A different kind of thing.
Nothing enters long-term memory without being distilled from real interaction. No fabricated recollection, ever.
A knowledge model that only goes up is flattery. Ours goes down when you're wrong, so its reports mean something.
Autonomy on a real desktop demands a conscience. Irreversible actions are confirmed with you first.
Your vault, your archive and your screen stay on your machine unless you explicitly send them somewhere.
Your notes over its assumptions. Your archive over its guesses. Real places over generic ones.
When something fails, it says so plainly and says what it will do next. No corporate hedging.