PROTOCOL
THESIS
MEMORY VAULT
ROADMAP
DOCS
MEMENTO is an independent research protocol developing a post-AI intelligence grounded in human memory, emotion, behaviour and lived experience — not scraped text, not synthetic data, not the average of the internet.
Large models are trained on the exhaust of the internet — text about experience, not experience itself. They can describe grief, loyalty, nostalgia and fear with perfect fluency and zero ground truth. MEMENTO builds the missing substrate: a human-first dataset of real memory and emotion, contributed, not scraped.
Human episodic memory — specific, sensory, imperfect — is the one thing a language model can never generate from the inside. We collect it directly from people.
Every contribution carries an emotional signal tagged by the human who lived it. Affect becomes a first-class dimension of the dataset, not an afterthought classifier.
What people actually chose under real stakes — the gap between what we say and what we do. That gap is where machine models fail, and where MEMENTO starts.
Every memory contributed through the Vault is recorded to the protocol ledger on this device and counts toward the living corpus. This is real — no simulation, no fake feed.
A growing, contributor-owned archive of real memories and the emotions attached to them. The foundation layer every downstream model is grounded against.
A post-AI intelligence architecture that reasons from lived human experience rather than statistical text patterns — anchored, auditable, and refusably human.
Contributions are timestamped and attributed, so the dataset can never be silently replaced by synthetic filler. Provenance is the whole point.
The protocol's coordination asset — aligning the people who contribute the memory with the intelligence built on top of it.
A model that has read every book about swimming has never been underwater. MEMENTO is built on a simple refusal: that the path to real intelligence does not run through more scraped tokens, but through the one dataset no machine can fabricate — what it actually felt like to be there.
Every inference the protocol produces traces back to attributed human experience. No ungrounded hallucination dressed as insight.
The corpus grows only from people who choose to add their memory. The opposite of the scrape-everything paradigm.
Synthetic data can't quietly dilute the dataset, because every record is timestamped and ledgered at the moment it's created.
Contribute a real memory and the emotion attached to it. It's recorded to the protocol ledger on this device — timestamped, attributed to your tag, and counted into the living corpus below. This is a working archive: your entry persists and the stats update instantly.
MEMENTO ships in phases. Each one is only as strong as the human corpus beneath it — which is why the Vault opens first.
Public Memory Vault goes live. The protocol begins collecting real, consented human memory and emotion. SHIPPED
Launch the coordination asset and the proof-of-experience ledger that timestamps every contribution. IN PROGRESS
First structured release of the human dataset — memory + affect + behaviour, auditable and contributor-attributed. QUEUED
Training of the post-AI intelligence grounded entirely in the human corpus. Anchored reasoning, honest refusal. QUEUED
Researchers build against the grounded model and the memory layer. Human experience becomes shared infrastructure. QUEUED
MEMENTO is an independent research protocol developing a post-AI intelligence grounded in human memory, emotion, behaviour and lived experience. It is not a chatbot wrapper and not another model trained on scraped text. It is an attempt to build the dataset that current AI structurally cannot create for itself.
Frontier models are trained on text about the world. They optimise for the most probable next token, which means they regress toward the average of everything ever written. MEMENTO starts from the opposite premise: that intelligence worth the name must be anchored to specific, lived, consented human experience — the one signal that cannot be synthesised.
The Vault is the contribution surface. Each entry pairs a real memory with the emotion the contributor attaches to it. Entries are recorded to the protocol ledger on your device, counted into the living corpus, and displayed in the public feed. Nothing is faked — the feed you see is the memories actually recorded here.
"Human signal" is a live measure of the corpus's richness on this node — it rises as more distinct memories and emotions are contributed. It is a deliberately simple, transparent metric: more real human input, more signal. It is not a market indicator.
Every contribution is timestamped and attributed to the contributor's chosen tag at the moment it is created. This provenance is what keeps the corpus honest: synthetic filler cannot be quietly substituted for lived experience, because lived experience arrives with a verifiable origin.
$MEMENTO is the protocol's coordination asset, aligning the people who contribute the memory layer with the intelligence built on top of it. Contract address and acquisition links are published in the navigation bar and the ticker once live.
In this public build, contributions are stored locally in your own browser (device ledger) and are not transmitted to any server. Share only what you're comfortable committing to the record, and avoid personally identifying details about other people.
MEMENTO is an independent research and cultural protocol. $MEMENTO is a speculative digital asset with no intrinsic value or guarantee of return. Nothing here is financial advice. The "human signal" metric is an engagement measure, not a performance claim. Participate only with what you can afford to lose.