Mei's inference APIs read a user's own words and return their likely age — and other demographic and behavioral signals — so your AI can adapt to the actual person, the way a human would. Not just for compliance. For AI that finally has EQ.
Today's LLMs speak the same way to everyone — a 15-year-old and a 50-year-old get the identical voice — and they were trained on the open internet, not on how people actually talk to each other.
A human rep, coach, or friend instantly reads who they're talking to and adapts their tone, examples, and pacing. Your AI can't — unless you give it the signal. And the model can only work with what's in the prompt; everything it could infer about the person is left on the table.
Word choice, phrasing, typos, emoji, timing — every message carries signal about who's writing it: their age, their mood, whether this is a romantic, professional, or family conversation. People can't consciously process all of it in real time. Models can.
We've spent a decade turning that signal into predictions — trained on 1.59 billion real, consented messages, the one corpus built on how humans actually talk to each other, not scraped forums.
One API call on a snippet of the user's writing. Age is live and self-serve today; the same engine infers the rest for enterprise clients on request.
Over/under-age verdict + age range from a text sample.
Explore the APIInferred from writing style.
Romantic, friendly, professional, family.
How the person is feeling, in the moment and over time.
What's on their mind, from what they write about.
Traits like trusting vs. guarded, direct vs. tentative.
Age assurance for Ofcom, COPPA, and KOSA is the obvious reason to infer a user's age — and a real one. But it's the smaller opportunity.
The bigger one is an AI that tailors its tone, examples, pacing, and guidance to the person in front of it. The same signal that flags a minor lets a companion app, a coach, a support bot, or a tutor meet each user where they actually are — the difference between a chatbot and something that feels like it gets you.
Pass a snippet of the user's own writing to the API — a message, a few turns of chat.
Receive the inferred signals as JSON — age today, plus whatever else you've enabled.
Fold them into your system prompt or routing. No model to train, no PII required.
Inference from text alone — no photo, no ID, no PII exchanged.
Age is live today as a self-serve API. Tell us what else you'd want to infer about your users, and we'll wire it up.