The Crier in the Square: Why We Built OYEE, a Free AI-Literacy Journey That Never Phones Home

Last month in these pages I argued that AI literacy is the new digital divide — that the real gap is no longer who has the tools, but who can question them. An argument is cheap. So this month, instead of another one, I am bringing you what my team built: OYEE, a free, bilingual AI-literacy journey at oyee.ca, for anyone fourteen and up. It runs entirely in your browser. It asks for no account, sets no trackers, and transmits nothing. And before you read further, the disclosure: OYEE is an Inquisitive Flow project, and Agora is its editorial arm. A publication that teaches doubt should begin by earning yours.

A Crier for This Square

Agora is named for the Greek public square — the place where a city argued with itself. OYEE is named for the sound that opened that square's business: the town crier's call, « Oyez! Oyez! » — hear ye. The crier had one job that mattered: to walk to the middle of the square and bring news people could trust, out loud, to everyone at once, whether or not they could read. The kinship between this publication and that site is deliberate. One is the square; the other is the voice in it.

Let me say plainly what the site itself says on its own pages: OYEE was built by a small team of four educators from the Inquisitive Flow family in Montreal — the same family that builds the Mnemosyne tutor and publishes this magazine. We are not a neutral observer reviewing someone else's product. This is our work, presented as our work. I have spent enough columns criticizing undisclosed conflicts in Canadian AI discourse that I am not going to manufacture one here.

Seven Chapters, One Town

OYEE's free journey is seven chapters and a capstone, carried by one continuous story rather than a syllabus. There is a town. There is a public square. A machine arrives, and rumours start — about what it knows, what it steals, what it replaces — and the learner's job, chapter by chapter, is to investigate them. We chose narrative over modules for a simple reason: nobody drops out of a story they are inside of.

The chapters are things you do, not things you watch. You train a small creature and watch it inherit the biases of its data — you do not read about bias, you watch it happen to something you trained yourself. You play a hallucination game across ten verified rounds, learning to catch a confident machine in the act of inventing. You watch a next-word predictor assemble sentences one guess at a time, which quietly dissolves the most stubborn myth in the room: that the machine knows things.

Then the journey turns to where your words go. A set of Law 25 scenarios walks through what actually happens when someone in Quebec types into a chatbot — which borders the words cross, which rights follow them [1]. Another chapter puts an answer-machine and a real tutor side by side on the same homework and asks the learner to judge the difference. The journey ends with a twelve-question capstone and a printable attestation — more on that document below, because how it is made matters as much as what it says.

Deeper Rooms for the Curious

Not everyone stops at literacy; some people want to see the machine room. So behind the journey there are deeper rooms: a step-by-step visual walk through a model's inner route — tokens, embeddings, attention, temperature — each stage shown on screen, not described in a paragraph.

And then the labs, the part I am proudest of as a builder: a real attention heatmap you can probe, a gradient-descent surface you can actually descend, a sampling slider that lets you feel what temperature does to a sentence. All of it hand-built on the canvas element, with no libraries. That was not a flex. A site that teaches people to ask what is inside the machine should itself be inspectable — view the source, and it is all there.

Nothing Leaves the Device

Now the part I consider the most radical, though it looks like a footnote: the whole site runs in your browser. No account. No tracking. No analytics. Nothing you type is stored anywhere or transmitted to anyone — not minimized, not anonymized, not shared with trusted partners: absent. The site says it in one line: « Rien ne quitte l'appareil, rien ne quitte le pays. » Nothing leaves the device, nothing leaves the country. I have argued here more than once that data sovereignty is decided in architecture, not in policy documents. This is sovereignty practised, not preached.

It costs us something real. We cannot tell you how many people have finished the journey, because we refuse to count. The printable attestation writes the learner's name into the page and nowhere else — no registry, no verification database, no record that it was ever issued. If that makes the attestation worth less to an HR department, so be it. It was designed for the person who earned it, not for a database.

Doubt as a Skill

The pedagogy runs on one rule we call live-before-explain: touch the mechanism first, name it after. You train the creature before anyone says “training data.” You watch the predictor guess before anyone says “language model.” Doubt is taught as a skill — with repetitions and feedback — not as a mood of general suspicion. And the journey never sells anything mid-lesson: no upgrade prompt, no locked chapter, no email capture standing between a learner and the next idea. It is the same order UNESCO's competency frameworks prescribe: understanding before operation [2].

None of this rests on instinct alone; the design leans on evidence the site itself cites. Bastani and colleagues showed in a large randomized study that unrestricted chatbot access can act as a crutch — students perform better with it and worse once it is taken away [3]. Kestin and colleagues showed the other face of the same coin: an AI tutor deliberately built on active-learning principles outperformed the equivalent in-class session [4]. Design decides the outcome, in both directions. That is why one OYEE chapter is literally a side-by-side of an answer-machine and a tutor: the difference between those two studies, made touchable.

The Canadian Frame

The timing is not an accident. In June, Canada's new national AI strategy — AI for All — put AI literacy at national scale on paper as a stated goal [5]. Quebec got there earlier: the Ministère de l'Éducation's digital-competency and AI guidance in 2024, then the Ministère de l'Enseignement supérieur's framework for higher education in 2025 [6]. OYEE aligns with those frameworks deliberately — French first, Law 25 woven into the story, doubt before operation. Aligned is the precise word. No ministry has endorsed this project, and none was asked to. A free journey does not need permission to exist.

This is what I meant last month when I wrote that AI literacy should be a public mission, not a premium product. A mission is not a sentiment; it is a deliverable. Free, in both languages, French first, for a fourteen-year-old in Rimouski and her grandmother in the same evening, on a school Chromebook or a library terminal, with no account standing in the way. That is the crier's deal with the square: the news is for everyone, or it is not news.

How It Stays Free

One section of honesty before I close, because this is where announcements usually get slippery. OYEE has a paid side. The same four educators deliver bilingual workshops and training for schools and organizations — and the site itself is the courseware, so there are no slide decks; participants work through the same rooms you can visit tonight for free. Requests get a quote within forty-eight hours. That revenue funds the free journey, and it is the entire model: no ads, no data resale, no premium tier carved out of the chapters. The lessons never sell; the selling happens here, in one paragraph, clearly labelled.

The crier did not follow people home to see what they did with the news, and he did not charge admission at the edge of the square. He called out what he knew, to everyone, and trusted the town to argue about it. That felt like a good constitution for an AI-literacy project in 2026. The square is open. Oyez.