Frequently asked questions

These are questions we’ve been hearing as we start talking with developers outside the Collective.

About the toolkit

What is the toolkit? A set of free, open-source building blocks that developers add to their own apps, so an app can learn what a person needs and, with their permission, adapt to those needs. It’s a code library, and apps can also connect to an optional service. How the building blocks work

Is it ready to use? The toolkit is still an early research version. Developers and community members can explore it, give feedback, and contribute, but using it today requires some prior technical development experience. We also haven’t yet rigorously tested example adapters with the communities they’re meant to serve.

Is this an accessibility overlay? No. A website owner chooses an overlay, usually to meet accessibility requirements, and often the person using the website still has to adapt to it. With the toolkit, the person chooses every change. We chose not to package the adapters as a compliance product, and the toolkit can’t make up for accessibility work a site hasn’t done.

Does it replace my screen reader or other assistive technology? No. We built it to work alongside what you already use.

Who made it? The AI for Accessibility Collective: Stanford HAI, MIT Media Lab, the University of Washington CREATE, and UCL with the Global Disability Innovation Hub, with support from Google.org. Meet the Collective

What did people tell you early on? People in our first sessions didn’t want to be filed under a category they had to pick, so the toolkit starts from what you say is hard to access. They also wanted to see what had changed and be able to take it back, which is why each change is visible and can be undone.

What could the toolkit become? We haven’t decided, but a few ideas keep coming up:

  • An adapter marketplace, where developers and users publish adapters and reuse each other’s, so a change someone builds once is there for everyone.
  • An open standard for profiles: a common format any app can read, so your preferences work across any app you want to use.
  • A home for the profiles themselves, ranging from a central service to a profile each person carries.
  • A way for people to build for themselves, where people with disabilities describe a need in plain words and AI coding tools help them build it.

Your profile and your privacy

Do I have to say I’m disabled? No. Your profile records what you ask for, such as larger text, not a condition. Plenty of people choose not to identify as disabled when they sign up for something, and the toolkit doesn’t ask you to.

Can the toolkit help someone with more than one disability, such as someone who is deaf and has low vision? Yes. The Ability profile agent focuses on what a person asks for, so one profile can hold several needs. We haven’t yet tried the toolkit for a combination of ability profiles, though, so if you build something for this, we’d love to hear any feedback or suggestions for improvement.

Do I stay in control? Yes. You can undo any change or add, change, or delete anything the profile contains. When an AI agent reaches a decision that’s yours, such as paying, it stops and asks you.

Who can see my profile? Only the apps you allow, and you can change that permission anytime.

Where does my profile live, and can it follow me between apps? We haven’t settled that yet. Today a profile can either stay inside one app, with no outside service, or live in a separate service that follows you across the apps you allow, with permission set app by app. Later, the choice could run from a central service to a profile you carry yourself.

Does the toolkit remember what I do on sensitive sites? Not by default. It does not remember from pages about banking, health and government services, so it doesn’t learn from them.

Will my information be shared with anyone else? Apps read only what you allow.

Can I set up my profile if I don’t speak, or if speech software struggles to understand me? Yes. You can type what you need instead of saying it out loud. The toolkit can’t yet understand speech that most speech software finds hard to recognize, though, and we’d like to add that in the future.

Does the toolkit learn from me, or is it a fixed list of settings? It can learn from you, if the app you are using incorporates the Ability profile agent, and if you agree to have it save profile settings. Your profile records the changes you agree to, so an app can adapt to you over time.

For developers

Why would my product use it? It’s the question developers ask us most. The possible benefits are adaptations you can reuse across apps, people not having to set their preferences each time when they visit an app or a website. We still have to show them in real products, and we’d like to do that with you.

How is a toolkit different from a product? A product does a single job. The toolkit offers building blocks to people building accessibility tools and to those building AI agents that are accessible to everyone, so a product can use only the pieces it needs.

Can I use just one part? Yes. You can use single adapters or verification checks without a profile, or only the voice and text control.

Can I use my own AI model? Yes, or even without an AI model. The structural parts work without a model, and you connect whichever model provider you already use.

Does it cost anything? The toolkit is free and open source. Some features use an AI model, which whoever runs the app connects and pays for. The personalized browser extension needs your own API key, and asks for it the first time you launch it.

What does the first version cover? Mostly adapters for a browser extension, and a few early ones for assistive wearables.

Are the adapters that come with the toolkit finished and tested? Not yet. They’re examples to start from, so please feel free to adjust them, test them or build your own.

What if an adaptation breaks a page? Every adaptation can be turned off, which returns the page to its original state. Some adaptations automatically check their own changes while they run. For example, Big Targets shrinks a button back to its original size if enlarging it would make its container overflow or push it past the edge of the page. Changes to your settings can also be undone one at a time, starting from the most recent, or reset to your profile.

How do you know an adaptation actually helps? Some adaptations check their own changes while they run, and others are tied to the Verification agent. However, whether a change helps is personal to the user, and we don’t currently have a way to evaluate this.

Does it work with screen readers? We designed it to work alongside assistive technology. It’s still at an early stage, and needs testing with a range of devices, including screen readers.

Taking part

I’m not technical. Can I take part? Yes. Tell us about a problem that gets in your way, try something and tell us how it went, or test what other people build. Take part without writing code

How do I report a problem or give feedback? Tell us through our contact form, or email ai4a11y-collective-contact@googlegroups.com. Developers can also report it on the toolkit’s GitHub page. Please report security problems privately to ai4a11y-collective-contact@googlegroups.com, never in a public issue. Only the Collective’s PMO sees what you send, unless they need to bring in someone else to deal with it.