keybr.com: Adaptive Touch Typing Trainer with Per-Key Statistics

What Makes keybr.com Different from a Standard Typing Test

Standard typing tests present a fixed text and measure words per minute. keybr.com takes a different approach: it tracks every individual keystroke and builds per-key statistics for each user. From those statistics, it automatically generates lessons that emphasise the letters and bigrams where the user makes the most errors or types most slowly. The README describes this feedback loop as the core differentiator: 'keybr.com offers a few unique features.' The system starts with a small set of the most frequent letters in the user's selected language, then adds new letters one by one as the user reaches the target speed with the current set. This staged expansion means a new user is never overwhelmed by the full alphabet at once. The README also describes a prediction feature: the system estimates how many more lessons the user will need to complete before reaching their configured target speed, which gives learners a concrete milestone to work toward rather than an open-ended practice session.

Per-Key Statistics and the Lesson Generation Algorithm

The statistical backbone of keybr.com is its per-keystroke tracking. For every key the user presses, the system records whether the key was correct, how long it took, and the context of the surrounding characters. These metrics accumulate over sessions and form the basis for the lesson generator's decisions about which letters to include in the next practice text. Letters where accuracy is consistently low or speed is consistently below the target receive more exposure in subsequent lessons. The README describes an automatic expansion mechanism: once a user reaches the target speed on the current letter set, a new letter is added. This means the practice text is never a generic sample sentence but a specifically constructed sequence optimised for that user's current weaknesses. The profile page, which the README describes as offering detailed graphs of learning progress, visualises this data over time and shows the improvement curve for each key.

Self-Hosting with Docker

The repository includes a Dockerfile and docker-compose.yaml for self-hosting. The Dockerfile builds the application in a Node 26 image:

npm ci
npm run compile && npm run build

The docker-compose.yaml exposes port 3000 and requires two volume mounts: a data directory for persistent storage and a .env file at a fixed path inside the container:

services:
  keybr:
    build:
      dockerfile: ./Dockerfile
    volumes:
      - /path/to/data_dir:~/.local/state/keybr
      - /path/to/data_dir/.env:/etc/keybr/env
    ports:
      - 30044:3000
    restart: unless-stopped

The .env.example file documents the required environment variables. The minimal configuration sets `APP_URL`, `COOKIE_DOMAIN`, `COOKIE_SECURE`, and `DATA_DIR`. The database defaults to SQLite with a local file path:

DATABASE_CLIENT=sqlite
DATABASE_FILENAME=~/.local/state/keybr/database.sqlite

Mail configuration is optional but required for email-based authentication flows. For local development without Docker, the package.json start script is `npm run start`, and `npm run start-docker` is the production path that initialises the database before starting the server.

Repository Structure and Technology Stack

keybr.com is a TypeScript monorepo managed with npm workspaces. The package.json lists the workspace root and two workspace groups: packages/* and scripts. Build tooling uses Webpack for bundling, Lage for task orchestration across the monorepo, and ESLint and Stylelint for linting. The HTTP server uses a custom framework called @fastr/core, visible in the package.json dependencies, rather than Express or Fastify. The packages/ directory contains the individual modules of the application: each key metric calculation, lesson generator component, and UI component lives in its own package. The repository also includes documentation in docs/ covering how to add a custom keyboard layout and how to add a new language. Both docs are relevant to contributors who want to extend coverage: the README explicitly calls out keyboard and language additions as a contribution path open to the community.

Licence Implications for Self-Hosting and Modification

keybr.com is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). The AGPL differs from the standard GPL in a specific and consequential way: it requires that anyone who deploys the software as a network service make the source code of any modifications available to users of that service. For a self-hoster running the application privately without modifications, this is not a concern. For a company that modifies keybr.com and deploys it as a publicly accessible service, AGPL-3.0 requires releasing those modifications. This is a hard constraint for operators who want to build a proprietary product on top of the keybr.com codebase. The hosted version at www.keybr.com runs the upstream code under these terms.

Limitations and What the README Does Not Cover

The README does not document rollback procedures for the database, which matters for self-hosters who want to apply or revert migrations safely. It also does not describe what happens when a user clears their browser data: since the self-hosted version uses a database for user accounts, data should persist across sessions, but the README does not explicitly confirm this for all storage configurations. The keybr.com lesson generator is optimised for Roman-script keyboard layouts; the README describes a process for adding custom keyboard layouts and languages, but this requires contributing to the repository rather than being a runtime configuration option. The application does not document a REST API for programmatic access to per-key statistics, so integrating keybr.com data into an external dashboard would require reading directly from the SQLite database. The AGPL-3.0 licence is worth noting for commercial operators: any public deployment of a modified version of keybr.com must release the modified source to users of that service. The monorepo structure, with packages/ containing individual module packages and scripts/ for build utilities, means a contributor needs to understand the Lage task runner and npm workspaces to build and test a single package in isolation.

Editorial conclusion

keybr.com is a solid choice for anyone serious about improving touch typing speed who wants a tool that adjusts its lessons based on their actual weak keys rather than presenting a fixed curriculum. It is also the right choice for self-hosters who want to run the full application with their own database and without sharing data with a third party. The AGPL-3.0 licence requires that any public deployment of a modified version release the source, which is a constraint for commercial operators. Before self-hosting, verify the Node 26 requirement and set up a SQLite or compatible database using the variables in .env.example.

Frequently asked questions

Official sources

1. aradzie/keybr.com on GitHub
2. Issues
3. License: AGPL-3.0
4. Project website
5. README