Back to the demos

Text version of the video: A developer builds a reading app with the toolkit

About this video

A narrated screen recording, about 5 and a half minutes long. The narrator shows the AI for Accessibility Toolkit on GitHub, runs a small reading app built on it, and then walks through the tutorial in the app's README. The narration below is transcribed from the video and lightly tidied. Descriptions of what is on screen are in square brackets, with the time in the video where each part starts.

Part 1: What the toolkit is (0:00)

[On screen: the AI for Accessibility Collective's page on GitHub. It lists four repositories: the toolkit, the browser extension, the repository for smart glasses, and the reader demo.]

Narrator: This is the AI for Accessibility Toolkit. It's a library that applications can embed to adapt themselves to every person using them. Instead of every application storing its own pile of settings, the toolkit keeps one profile of what one person needs: for example, bigger text, more line spacing, a dark theme. Each app renders that profile into its own adjustments. The vision of the toolkit is to onboard once, and then every application that embeds it can follow and be accessible to users, no matter what their needs. Everything is open source under the AI for Accessibility Collective repository.

Part 2: The toolkit's repository (0:40)

[On screen: the toolkit's repository on GitHub. The narrator scrolls through its folders and down to its README, titled "AI for Accessibility Toolkit: an open toolkit for building apps that adapt to each person's accessibility needs."]

Narrator: This is the toolkit itself. The core of it holds the profile and a catalog of about 50 adapters that can make accessibility fixes on a page, and a hosted service for apps that aren't written in JavaScript.

Part 3: The reading app (1:05)

[On screen: the reader demo's repository on GitHub, described as "a one-page reading app that adapts to the person: the smallest complete host of the AI for Accessibility Toolkit, written as a tutorial".]

Narrator: This is the repository for the AI for Accessibility Reader demo, a starter application.

[On screen: the reading app open in a web browser. At the top is a text box that says "Tell me what you need", with a Speak button, a Send button and a checkbox labeled "Speak replies", which is ticked. Below the box, a line of help text says you can type or say what you need, such as bigger text, more space between lines, dark mode, a dyslexia font, or a description of yourself like "I have low vision", and that you can ask the app to read the article aloud or ask questions about it. A status line says "Profile: nothing yet". Below that is a short essay titled "Why reading is work, and what a page can do about it".]

Narrator: This is the application itself, the smallest complete app built on the toolkit. Its README is a 10-minute tutorial, but let me show the application first. It's one article and one input box at the top.

[The narrator types "make the text bigger" and sends it. The app replies "Making the text bigger", the essay's text grows, and the profile line now shows a text size of 1.2.]

Narrator: So I'll type in here, make the text bigger. The text then grows.

[The narrator types "I'm dyslexic" and sends it. The app replies "Setting up support for dyslexia". The essay changes to the app's dyslexia font, with more space between the lines. The profile line now shows text size 1.2, the dyslexia font turned on, and line spacing 1.8.]

Narrator: I'm dyslexic. Now the dyslexia font is being used, with wider line spacing. The page has recorded that as a profile, not purely as a font size.

[The narrator types a question asking what the last paragraph of the article says. The app answers in writing at the top of the page and reads its answer aloud: "The last paragraph explains how this demo allows you to state your reading needs once by typing or speaking. The page remembers your profile and automatically applies adjustments whenever you return or ask for changes. It concludes that the page itself should do the work of adapting to the reader."]

Narrator: This answers questions about the article, for instance for someone who might not be able to see it. That same command could also be given by clicking Speak.

[The narrator points to the Speak button, then reloads the page. The page comes back already adapted: the profile line still shows text size 1.2, the dyslexia font and line spacing 1.8.]

Part 4: How an app uses the toolkit (3:06)

[On screen: the reader demo's README on GitHub, titled "Reader: learn the AI for Accessibility Toolkit in ten minutes". It says that to run the app you need Node 20.12 or later and a Gemini API key. A diagram shows four stages from left to right: what the person says, such as "bigger text" or "I'm dyslexic"; the profile, which records needs such as text size 1.2 and the dyslexia font; the settings the app works out from the profile, such as a font scale of 120; and the page, where adapters change what you see. The narrator scrolls down through three steps, each with a short code example.]

Narrator: So coming back to the reader repository: if you want to learn about the toolkit, here is what you can take a look at.

Step one: your application creates the toolkit and hands it the ports: storage, clock, scheduler. The core has no platform code in it, so this demo implements exactly one port, 29 lines over local storage, and the same core runs in a browser, in Node or on a phone.

Step two: your application needs to write the profile. For example, text size changing, dark theme changing, and that's all written through one object, the Librarian. That's the only thing that can write the profile.

Step three: your app then reads those settings back from the profile, rendering them on its surface. In the example we showed, this was the web, but it could be any application, be it mobile or extended reality. It feeds those settings to adapters it gets from the catalog. Three adapters, a dozen lines: the app itself never wrote a fix.

So where did the needs come from? Here, a model turns what the person says into a small vocabulary of needs, and the server checks it before anything is written. That's just this app's choice, however. Settings could work any way the developer configures.

So that's the toolkit: a profile of needs, rendered by each app, applied by adapters.

Part 5: Try it yourself (4:50)

[On screen: the reading app again, then the reader demo's repository, then the Collective's page on GitHub, and finally the toolkit's repository, where the video ends.]

Narrator: To learn more, please do try the demo, the reader application, by pointing your AI coding tool at it. Then run it, and ask questions through your coding tool. When you're done, switch over to a new branch of your own application. From your own application, ask your coding agent to add agentic accessibility, and point it at this toolkit to do so. Please try this out and let us know your feedback. Thank you.

[End of video, at 5:32.]

Back to the demos