KeyRecall is an adaptive practice partner for piano technique. Connect a MIDI keyboard, play the scale it gives you, and it decides what you play next: which key, which form, how many octaves, hands separately or together, at what tempo, and how much of the notation to show you. It watches what you actually played and keeps a long-running picture of what you can do and what you are starting to forget.
The interaction is meant to be almost nothing:
Open it. Play what it gives you. Stop whenever you want.
Everything else is underneath.
-
It chooses, so you don't
No lesson list to pick from and no streak to maintain. Each exercise is chosen from what you have played, how long ago, and how it went, aiming at the edge of what you can currently do rather than at whatever comes next in a book. -
It watches you play
Every note you play is matched against the scale that was asked for. Playing the wrong notes, stopping and starting, playing unevenly, and hands that come apart are four different problems, and KeyRecall keeps them apart instead of collapsing them into one score. -
Support that fades
A scale you do not know yet is shown on the staff while you play it. Once you can, the notes are previewed and then hidden. Once that is comfortable, you are asked for it from memory. Moving down a rung is a normal response to a bad attempt, not a penalty. -
Remembering is the point
Playing a scale with the notes in front of you is practice, not recall, and KeyRecall records it as such. What moves your memory of a key is producing it without being shown, and the schedule is built around bringing each key back as its recall starts to weaken. -
A count-in, and a metronome only if you want one
Every exercise counts you in, at every level of support. The metronome is a separate choice you make, not a reward you unlock. -
One install, several people
Each profile keeps its own history, and switching between them takes a tap. A household piano is usually not one person's. -
Your practice history is yours
Everything is stored on your device as an append-only record of what you played. Nothing is uploaded, there are no accounts, and there is no advertising or tracking. Your skill picture is recomputed from that record rather than stored as a score somebody could quietly rewrite.
All twelve keys in four forms: major, natural minor, harmonic minor, and melodic minor. Each can be asked for with either hand alone or hands together, over one octave or two, ascending or ascending and descending, in parallel or contrary motion, at four tempos, and at any of the three levels of support.
That comes to 9,216 distinct exercises, which is the real argument for letting something else choose. A practice plan you write yourself fixes one path through that space in advance, and what you most need to play today is rarely the next line on it.
So the scope is narrow but not small. Scales alone hold years of progression, and doing one thing well is worth more than doing everything shallowly. Arpeggios are not here yet. They are the most likely thing to be added next.
- Pianists who know they should practice technique and would rather not also plan it
- Returning players with uneven ground behind them, who need the gaps found rather than guessed at
- Students working through scales systematically, who want the review timed rather than arbitrary
- Teachers curious about what an evidence-driven practice schedule looks like
It is a practice partner for technique, not a piano course. It has no repertoire, sight reading, ear training, or theory lessons, and it is not trying to acquire them.
KeyRecall is not released yet. It is under active development and there is nothing to install. The learner model and scheduler are built and documented and the practice loop runs. The structure underneath is settled; the constants in it are versioned starting points, calibrated against recorded playing where they could be and not yet against a real learner population.
This is a human-led project built with assistance from AI coding tools. AI tools may be used as part of the development workflow for exploration, implementation support, refactoring, testing ideas, and review, while real people remain responsible for design decisions, validation, maintenance, and project direction. The project is developed with deliberate attention to pedagogical honesty, musical correctness, accessibility, cross-platform behavior, and long-term maintainability.
The documentation map covers the product, the domain model, and the research record. For the learner model and scheduler as they currently stand, start with The KeyRecall V1 Adaptive System.
To build or work on KeyRecall, see CONTRIBUTING.md.
KeyRecall is released under the Zero Clause BSD License (SPDX: 0BSD).
Copyright © 2026 Aaron Bull Schaefer and contributors