The Shape of Fido's Software
Software in the service of a school
Fido's school is still being built, and so is its software. This article describes the software as it stands in November 2026: what it is for, the commitments behind it, its parts, what we build ourselves, and what is still missing. It is written for parents, educators, funders, and anyone who might help build it.
It opens a second part of the Fido model, on how the rest of Fido works. It assumes you know the model in outline; The Fido Model at a Glance gives it in ten minutes. How we build the school in general, from the loop with real children to the use of AI, is told in How We Build the School. This article does not repeat it.
One rule runs through every section. We say what runs today, what is planned, and what is still undecided. Where something is only a plan, we call it a plan.
What the Software Is For
A day at Fido has three modes, described in
A Day at Fido. In the mastery mode, each
child practices on a tablet at their own level. In the guided mode,
children work in workshops with a guide. In the free mode, they pursue
work and projects of their own. The software carries the first mode.
Its center is the
That sounds like any learning app. The difference lies in what has to work together. The curriculum is a knowledge graph, in which every topic knows what it builds on. A topic counts as mastered only once a child solves it reliably. Mastered topics return at growing intervals, which is called spaced repetition. Everything lines up with LehrplanPLUS, the Bavarian curriculum framework. And one view shows progress across all subjects. We found no product that combines these five, so we write our own.
The software does not replace the adults. It takes over the practice, so that the guide has time to explain, listen, and encourage. Why we build the content, the software, and the pedagogy from a single source is explained in How We Build the School.
The plan for the software is much larger than what runs today. The parts for children run and are used when children test our material. Much of what guides, parents, and the school office will need exists only as a design or as a demonstration; the list of what is missing names each.
Five Commitments and What They Cost
Every technical decision at Fido answers to a short list of commitments. When one of them blocks a path, we change the design, not the commitment. Five of them shape the daily work, and each has a price.
We own the core. The mastery engine, the knowledge graph, the record of what each child has learned, and the code that displays exercises are ours. Everything else may be borrowed. This costs us complexity. We accept it because a school lasts for decades, and a supplier who changes its terms should not be able to change what children practice.
Offline first. A school's Wi-Fi sometimes drops. Practice, reading, and progress then carry on, and results wait on the tablet until the connection returns. This rules out tools that must be online all the time. It is also why the app carries the curriculum with it instead of fetching every page.
Privacy by design. The GDPR, Europe's data protection law, is the floor, not the ceiling. We collect little, store it in the EU, and refuse surveillance signals, such as a camera that checks attention or typing speed turned into a "focus score". A class whose parents have not consented can use a local mode, in which the app sends no name, no answer, and no progress anywhere; the detail, including the crash reports the app sends in every mode, is in The App, the Server, and Where a Child's Work Lives. The price: some tools that would save us work are off the table.
One record, many views. What a child did is recorded
once, as an entry in an
Open code, private data. The software is meant to become a public good that other schools can use and adapt. What stays private is data, never code. How far this has got is told with what is still missing.
Three more commitments complete the list. Every screen is German first, with English fully supported. We use widely known tools that people and AI coding assistants both read well. And we add complexity only when the next step really needs it.
The Parts and How an Answer Travels
The software has a handful of parts, kept together in one code base.
- The curriculum holds every topic, exercise, and article as plain data files, not as screens. The other parts read it.
-
The shared engine is one library that displays
exercises and articles and contains the
mastery engine . It does not run on its own; the other parts build it in. - The app serves children, guides, parents, and the school office. It runs on Android tablets and in the browser.
- The server keeps the records, handles sign-in, and works out what each guide sees.
- The explorer is a public website on which anyone can read the whole curriculum, this article included.
- The website introduces the school.
How the parts fit together shows in the path of a single answer.
A child taps an answer. The exercise comes from the curriculum and is
drawn by the shared engine inside the app. The app writes the answer
into the tablet's own storage, as an entry in the
Three later articles take these parts further: Curriculum as Data, Rendered Anywhere, The Mastery Engine, and The App, the Server, and Where a Child's Work Lives.
What We Build and What We Choose
Three choices decide what the parts are made of, and each has a price.
One toolkit for tablet and browser. We compared five ways to build the app, from a plain web app to an app written only for Android. We chose Flutter, a toolkit that turns one code base into an Android app and a web app. It was the only option that did well on the three things that mattered most: speed and offline use on Android tablets, locking a tablet to the one app, and one code base for every role. The price: fewer developers know its language, Dart, and the web version starts more slowly than a lean website.
Paper-like screens for children. Children are to work on e-ink tablets, the screens known from e-readers. They are easy on the eyes, come with a pen for handwriting, and are too slow for videos and games. Guides are to get ordinary tablets, because their class overview has to respond at once. Parents and the school office use a browser. The price: e-ink tablets cost more than ordinary ones, and every exercise has to work on a screen that refreshes slowly. The app has an e-ink mode for this. Whether the exercises feel quick enough on e-ink, in children's hands, is the most important test still to run. It comes before any devices are bought.
Exercises of our own. The plan is to write most exercises ourselves, starting with mathematics and German, with AI drafting and teachers reviewing. Existing learning platforms may fill gaps for a while, where they fit the way children learn here. Today the software uses no outside platform at all. Every exercise comes from our own curriculum, which holds several thousand topics. The price is the writing: a curriculum of our own has to be written, checked, and kept up to date by us.
What Is Open, What Is Planned, What Is Missing
The software is meant to be open. The whole stack that runs a Fido school is to be published, so that other schools and researchers can use, change, and run it. The plan fixes the licenses: the MIT license for the code that displays content and for the explorer, CC BY-SA 4.0 for the curriculum, and AGPL-3.0 for the app, the server, and the mastery engine. Under the AGPL, anyone who runs a changed version as a service must publish the changes. What stays private is data: children's records, passwords, and keys.
None of it is published yet. The code lives in a private repository. The first step, making the display code usable on its own, is partly done. Publishing that code comes next, then the curriculum, the explorer, and finally the app and the server.
What runs today is mainly the child's path: articles, reading, practice, the mastery engine on the tablet, sign-in, and a class view for guides with real data, including a list of the children who need a look. Measured against the full plan, much is still missing.
- The parents' view of a child's progress, and the guide's tool for handing out Fido Bucks, run only in a demonstration mode, with sample data. Children already earn Fido Bucks by practicing; the Fido Store, where they are to spend them, is not built.
- Calls with outside subject tutors, the planning of workshops, and the curriculum report for the school authority are not built.
- Nor are the timing of the short practice blocks, notifications to parents, or a termly summary that could stand in for a report card.
Five more articles take the software further: Curriculum as Data, Rendered Anywhere, The Mastery Engine, The App, the Server, and Where a Child's Work Lives, How We Know It Works, and How a Small Team Builds Fido's Software with AI. Three more look beyond the software: Fido's Bet and How We Will Know, How Fido Is Meant to Reach Beyond One School, and How Fido Is Meant to Be Paid For. Two tell whom Fido serves: Who Fido Is For, and What It Owes and Fido Between Two Kinds of School.
Questions, objections, or keen to help build it? Write to us.