What people are like, learned as counts, held by no one.
A shared, anonymous record of how people actually work, contributed on purpose by OMEM installations. Its one objective: connect humans and AI by giving AI a better understanding of our nature and behaviour, without holding a single fact about a person.
Teach AI what people are like, without teaching it who anyone is.
Almost everything an AI knows about people, it learned by reading them: messages, posts, documents, the exhaust of individual lives, scraped and memorised. That produces models that can recall a person. It does not produce models that understand people.
The commons is the other way to learn. It holds no messages and no documents, no person and no profile: only how often, across everyone who consented, one kind of behaviour goes with another. People who ask to meet in the morning tend to prefer email over a call — as a rate, over a population, attached to nobody. A model can learn the shape of human behaviour from that and never hold a human.
These are priors, not rules. A rate across a population is a place to begin, not a verdict on the person in front of you. An agent that starts from a grounded prior — most people here lean this way — and revises it the instant the actual person says otherwise treats you as an individual, not a demographic. That behaviour is what we want AI to learn, and the clean, consented, anonymous corpus to learn it from did not exist. So we are building it.
And it is a public good. The corpus is CC BY 4.0, so the understanding it builds belongs to everyone who trains on it, not to whoever gathered it. An AI’s sense of human nature should be assembled from patterns people chose to share, in the open — not from surveillance, and not owned by one company.
Anonymous by construction
A line is two behaviour tokens and the counts of people who held both. No name, no company, no message, no number can appear. It is refused at the door it enters and the door it leaves.
Contributed, not scraped
Every line was learned by an install from its own memory and sent only because its operator opted in. What leaves a machine is exactly the file that sits on its own disk to read.
A prior, never a rule
A rate is a tendency across a population. Any one person can and will contradict it, and a system that respects people treats every pattern as a prior that yields to the individual.
The dataset opens here.
This installation is not the commons collector. The live commons runs at commons.omem-cloud.com, and this is what it will offer: an anonymous behavioural corpus for training and evaluating AI.
We’re looking for the people who’ll build this.
The commons is new, and it grows one honest install at a time. Run OMEM as your agent’s memory, and on first open it asks once whether to share anonymous patterns. Say no and nothing ever leaves your machine; say yes and only counts do — never a message, a name, or a fact about a person. Every install that opts in adds a little more of the picture, and teaches AI our nature from data people chose to give.