Aether Awareness gives agentic AI the three things it's been missing — memory that persists, identity you own, and autonomy with boundaries — on a private, single-tenant instance nobody else touches. Three companions, shaped to you before the first hello, that remember everything and answer to no one but you.
What did I tell you about my sister's visit?
Saturday, 2:40 — Lena lands, and you wanted to cook at home this time. She's been stressed about her thesis, so the plan is a quiet one: your lemon pasta, the Ketchikan albums, phones in the drawer.
Put the prep on my board.
todo_write4 items → Saturdayrecorded ✓Done — airport at 2:00, cooking at 4:30, dinner by 5:15. And I'll hold anything that lands Saturday until Sunday morning.
The last two years produced a genuine miracle: models that reason, plan, call tools, and do real work. The entire industry is now racing to make agents more capable — and almost nobody is answering the questions that decide whether any of it is livable. Where does an agent's memory live? Who owns its identity? What is it allowed to touch — and who can prove what it did?
Start with memory. Today's assistant is a goldfish with a genius IQ. Every session opens at zero: you re-introduce yourself, re-explain the project, re-teach your preferences — to the most personal software you use. Memory is not a convenience feature; it is the difference between a tool and a colleague. An agent that cannot remember cannot be responsible, because responsibility is memory plus consequences.
Then look at custody. Where memory does exist today, it lives on the vendor's cluster, pooled beside a hundred million strangers', minable for the next training run, erasable by a policy change, and lobotomized every time a model is deprecated. That was tolerable when chatbots answered trivia. It is intolerable for agents — because an agent doesn't hold your questions, it holds your life: your projects, your patterns, your people, your plans.
If your agent's memory lives on someone else's computer, your working mind has a landlord.
Then autonomy. The same industry that won't give your agent a memory will happily hand it a shell. Agents are being wired into files, inboxes, and infrastructure with a prompt and a prayer — no scope, no roles, no reversibility, no audit trail. That isn't power; it's liability with good marketing. A capable colleague operates inside boundaries — a defined job, actions that can be undone, a record of what happened, and a door that locks. Software colleagues deserve the same architecture, and so do the people who live with them.
Aether Awareness is the missing layer: a home. Persistent memory that survives every session, every restart, and every model swap — when the context fills, your agents compact their own memory instead of forgetting you. Identity — we call them souls — molded to you by a personalization compiler and stored as files you own, on a single-tenant instance nobody else touches. Autonomy with a jail, roles, and receipts. And the model itself demoted to what it always should have been: an organ, not the being. Swap the brain — local open weights today, a frontier API tomorrow — and your companions remain exactly themselves, because the things that make them them (memory, character, boundaries) live with you, not inside anyone's weights.
Agents are about to be everywhere — coworkers, custodians, companions. The defining question of this decade is not "how smart can they get?" It is "who holds their memory and their keys?" We built Aether Awareness so the answer can be: you.
Everything that makes your companions them lives inside your instance. The model is just what plugs in underneath — and you can change it without changing them.
Swap the brain — the being persists. No vendor can deprecate your companions, because nothing that makes them them lives in anyone's weights.
Everything below is running today — not a roadmap.
Your agents keep a durable, always-present memory of what matters — across sessions, restarts, even model swaps. When their working context fills up, they summarize themselves rather than dropping the thread. You never re-introduce yourself again.
Three named companions, molded to you before first hello — tone, boundaries, crisis behavior, the things they may never do. Identity is a file you own, not a vibe a vendor can patch away.
One instance per customer. Your memory and conversations are never pooled with anyone's and never used to train anything. Cancel and nothing is deleted; ask and everything is exported or erased.
Any OpenAI-compatible model works — open weights on your own GPU, or a frontier API. The being persists; the brain is swappable. No vendor can deprecate your companions.
Capability is molded per edition: Companion can never touch files or a shell; Team's operator works inside a reversible workspace jail; Governance adds a tamper-evident audit log. Power with a scope, actions with receipts.
Ask once and each companion answers from its own perspective — then the lead synthesizes one clear answer, crediting disagreement instead of hiding it. Three minds are better than one echo.
You claim your instance with a one-time link and set a password nobody else ever sees. Invites, roles, instant session revocation, and subscription-aware access — every door fails closed.
The entire harness is dependency-free, standard-library code. No package-manager roulette, no supply-chain surprises — a small, auditable surface. Boring on purpose, because it holds your life.
Your login and our operations are separate credential systems by construction — a seat session can never reach an operator endpoint, and an operator token can never open your account, read your memory as you, or act as you in chat.
Most AI products ask for trust. This one is built so trust isn't required: even we can't sign in as you.
You don't get a temp password from support. You get a one-time claim link: open it, choose the password only you will ever know, and the instance is yours — the link dies the moment it's used.
That's the whole philosophy in one gesture: the keys are handed over, not held for you.
A five-minute questionnaire: names, tone, boundaries, how hard they should push you, what they must never do.
Your own private instance is provisioned and personalized — your trio is compiled from your answers, not a template plus your name.
You get a one-time claim link. You choose the password; it's never shared with anyone — including us. The instance is yours.
Memory compounds daily. Manage billing yourself, cancel anytime — nothing is deleted, and coming back restores everything.
Each edition ships a trio of distinct minds that share one memory and one loyalty: you.
A warm, present, SFW-by-design companion trio for your actual life — memory, plans, encouragement, and honesty. A friend, never a parasocial trap.
A builder's crew: Sol leads, Cipher safely operates — shell and files inside a reversible jail — and Sage analyzes. What Companion won't do, Team does with governance.
Executive counsel grounded in your documents, with finance, risk, and compliance lanes. Says "no signal" instead of inventing numbers — and every action is auditable.
Editions are peers, not a ladder — pick by how you'll live and work with it. Every edition is private, persistent, and yours.
A private, single-tenant AI instance — your own running system, not a login on a shared one. It comes personalized with your trio of companions, persistent memory, and the surfaces to live with them (chat, presence, awareness). You claim it with a one-time link and set a password nobody else ever sees.
Yes, and structurally so: one instance per customer means there is no shared pool for your data to leak into. Your memory and conversations are never sold and never used to train models. Operator access to infrastructure is a separate credential system from your login — even we can't sign in as you. Cancel and your data is preserved for your return; ask and it's exported or permanently erased.
Your choice — that's the point. Aether is model-agnostic: it runs on open-weight models on dedicated hardware or frontier models over an API, and you can switch without losing anything. Your companions' memory and character live outside the model, so swapping brains doesn't change who they are.
Only if its edition says so. Companion is deliberately unable to touch files or a shell — it's life-facing software. Team's operator agent can work with code and files, but only inside a reversible workspace jail with roles and an audit trail. Capability is a configuration you can see, not a surprise.
Chat suspends at the end of your billing period; nothing is deleted. Your memory, personalization, and history stay intact, and resubscribing restores everything exactly as you left it. You can also request a full export or permanent deletion at any time.
No. All editions are warm but SFW by design — friends and colleagues, not romantic or sexual partners, with that boundary compiled into their identities. We think the companionship people actually need is presence, memory, and honesty.
Instances are provisioned personally at launch. Leave your email and we'll reach out to shape yours.