The memories that matter shouldn't fade.
MNEMO is a neurotech lab working toward a consumer EEG layer designed to tag the memories that matter most to you while you're awake, and reactivate them in the right phase of deep sleep — so the important ones are more likely to last. We're pre-validation, open about every gate we still have to pass, and building it in public.
The brain holds more than we can reach.
In 1885 Ebbinghaus measured how fast we forget — most of a new experience gone within a day. He used nonsense syllables. Autobiographical memory is far richer, and its loss far more costly.
When early Alzheimer's takes a grandparent's memory of a grandchild's first steps, something irreplaceable is erased. The trace of a memory is still there, written in the brain's electrical activity as we recall it — we just can't reliably reach it, hold it, or strengthen it.
MNEMO starts there: not to read minds, but to help the memories that matter most survive the night. That's a hard scientific problem, and we're early. The rest of this page is about exactly how early — and how we intend to earn every claim.
Working toward a headband that strengthens the memories that matter.
The idea: by day, MNEMO learns which of your memories carry weight; at night, it reactivates the important ones at the right moment in deep sleep; over time, it builds a private, searchable index of your own past. No prior work cues the specific memory that matters, chosen by salience — that gap is our wedge.
Salience, not surveillance
While you recall your day, MNEMO is designed to learn which memories matter to you — their salience and state — so the right ones can be strengthened. Not which thoughts you have; which ones count.
waking · salience & stateReactivation in deep sleep
Sleep is when memories consolidate. MNEMO aims to ease you into sleep and deepen slow-wave sleep, then replay a personal cue for your important memories, timed to the right phase of the slow wave. Getting that timing right on real hardware is one of our hardest unproven gates.
sleep · cue-specific TMRA memory vault that's yours
Your memories are meant to become a private, searchable index — in your own words — that grows more personal the more you use it. By design it lives with you, not on someone else's server.
private · personalisedThis is what we're building — our shipping goal, not a finished product. None of it is validated yet, the headband itself doesn't exist as hardware, and our first studies run on a lab EEG rig. Below, we're specific about which gates we have to pass before any of this is a claim.
What we ship — and where it leads.
We keep these strictly separate, on purpose. One is what we're working to ship and validate now. The other is a research horizon we'll only claim once the data earns it.
Strengthen the memories that matter
A consumer EEG layer for memory consolidation:
- Salience-tagged autobiographical memory
- Cue-specific reactivation in deep sleep (TMR)
- Easier sleep onset and deeper slow-wave sleep — our sleep goal
- A private memory vault that personalises each night
A richer portrait of a memory
As consented data accumulates, a longer-term research direction:
- A fuller "memory portrait" — image, words and feeling together
- Reconstructed from your own accumulated data, within-subject first
- A frontier we approach only through our own accumulated data
Not a promise, not near-term. We'll claim each step only once the evidence is in.
Honest about where we are — clear about why it works.
Being early isn't the same as being undifferentiated. Three things make this a defensible bet, even before the first gate.
Memory-specific cueing
Existing sleep-reactivation studies cue with generic sounds. None reactivate the specific memory that matters to you, chosen by salience. That space — affect-tagged, memory-specific reactivation — is the one no one occupies.
A dataset that compounds
Every consented night pairs salience and sleep with a next-day recall outcome. That federated affect-and-outcome dataset is the asset that grows with us — and that a competitor can't simply copy.
Parity is reachable; affect is at the threshold
Phase-aware sleep stimulation already ships in wearables; affect on consumer EEG is at the edge of feasibility. The bet is that the two — plus our wedge on top — become one product.
We don't have results yet. We have a plan to earn them — in public.
For decoding from low-channel EEG, the easiest thing in the world is to fool yourself. So our discipline is the product: falsifiable claims, decisions frozen before we look, and the negatives published alongside the positives.
Cross-subject affect-salience AUC ≥ 0.70 on 16 frontal channels (FACED, leave-one-subject-out) — beating subject-identity, ocular and amplitude baselines, permutation p < 0.01.
If it fails, we say so and pivot. The analysis is pre-registered before we run it and tested against a retrained-permutation null. We’ll never report a within-subject number as if it were cross-subject — and our earliest study is a small within-subject self-test: an engineering signal, not proof of efficacy. This is the first of several gates — deep-sleep specificity, on-hardware cue timing, and cued-vs-uncued recall each have their own pass/fail bar, in that order.
Decisions frozen up front
Metrics and thresholds go on OSF before data collection, with a reproducibility bundle on Zenodo.
Held-out people, not epochs
Every claim is measured on participants the model has never seen, with group-aware splits.
Beat the trivial explanation
Subject-ID, ocular and amplitude baselines run alongside. A neural result has to beat them or it isn't one.
Every advertised capability has its own kill-switch. We’re at gate 1 — the rest stay clearly marked “not yet” until their evidence is in.
You control your data — and contribution is always your choice.
- By default, processing runs on your device. Your raw recordings stay with you.
- If — and only if — you opt in, your device shares model improvements (gradients), never raw recordings, so the system gets sharper for you and for everyone who opts in.
- You choose your level: keep everything local, contribute to the shared model, or stop at any time.
- In a research study, data is processed by our team under your informed consent and ethics-board approval — a separate, explicitly opted-in mode, distinct from everyday on-device use.
- Consent is specific and revocable — nothing happens to your data without it, and you can withdraw at any time.
A memory tool only earns trust if you stay in control. And a model only keeps getting better with consented data from real use. We designed for both from the start — control first, contribution by choice — rather than promising your data never moves and quietly needing it to.
The work happens in the open — including the weeks it fails.
For an honest, pre-result company, openness is the strongest proof we can offer. Code, pre-registrations and weekly logs — so anyone can check our claims against our method.
A neuroscience-first team, not a black box.
Igor Zinoviev
CEO / Principal InvestigatorLeads the science and mission. Psychology and memory research, Maynooth University.
Danylo Shliahetko
CTO / Machine LearningBuilds the EEG models, closed-loop systems and validation harness. Lviv Polytechnic.
Andrii Vynnytskyi
Software / ML EngineerBuilds the data pipelines, on-device app and developer tooling.
Karolina Myrovska
COOOperations, data partnerships and research ethics. Ukrainian Catholic University.
We're assembling a scientific advisory board across memory science, sleep and neuroethics, and will name advisors here once they've agreed to be listed — no borrowed logos, no names we haven't earned.
Early signals you can check today: our founders' affiliations (Maynooth University, Lviv Polytechnic, Ukrainian Catholic University), our open code, and our NeuroTechX involvement.
Different ways in, depending on who you are.
We're early and honest about it. Here's how to engage based on what you're looking for.
Co-author the science
Memory, sleep or affective-neuroscience lab? We're looking for collaborators and co-authors as we move toward our first pre-registered study — the partnerships that make the science real.
research@mnemo.aiBack honest neurotech early
We're applying through deep-tech grants and accelerators and talking to investors who back falsifiable, mission-led science. Start with our research plan and the wedge above — then let's talk.
hello@mnemo.aiFollow & contribute
Read the weekly logs, check our method, try the code. The fastest way to understand MNEMO is to watch us work.
@mnemo_researchEarly design partners
Need affect or sleep labels on EEG for your own research? We take a small number of early design-partner / co-development engagements — we won't sell you a "validated" result we haven't earned yet.
partners@mnemo.aiThe boundaries are part of the product.
We'd rather lose a slide than a reviewer's trust. Here's where the lines are, plainly.
What MNEMO is
- A memory-consolidation tool: strengthen the memories you choose.
- Salience- and state-aware — which memories matter to you.
- Built to a falsifiable, pre-registered standard, in the open.
- Yours: on-device by default, consented contribution by choice.
What MNEMO won't pretend
- We don't reconstruct your memories, images or thoughts — that's a research horizon, not a product.
- We read salience and state, not emotion as a label or identity — and we don't read minds.
- Sleep stimulation is a research direction under ethics review with strict safety limits — not a service we sell.
- Research-grade and pre-clinical today — not a medical device, and never a result before its gate.