Animelon's way of learning from video, from the person's own Jellyfin server: connected per device (Settings, or a studyset's Add material › Watch from Jellyfin) with a name and password or a Quick Connect code, over HTTPS only (plain HTTP only to a server on the same machine); the token stays in the system's keyring. Shows, films and videos tied to a studyset follow the person in their settings as the server's and the item's ids alone (settings schema 9), named by the server when shown. The Watch screen plays the video (video_player on Android, media_kit's libmpv on Linux and Windows) with the line in the studyset's front language cut into the dictionary's words with furigana, and the back language's line beneath; a word tapped pauses the video and becomes a card with the line and its translation as its first hint, tagged with the show. The video's words are listed, most frequent first. The sound is the track in the studyset's language, whichever the file marks default, and the menu switches it; the subtitles read are a whole track, never a forced one (which only covers what the sound doesn't say), and captions for the hard of hearing only without plain ones. The line and its translation take the video's width in a box of a fixed height, so the video keeps its size; in the dark theme the video's colours glow around it, as YouTube's ambient mode, and the words' panel takes the colour of the episode's picture under the pointer. A line is cut into words with one request: DictionaryEndpoint.withKeys looks up every stretch of it that may be a word, as it stands and as every word it may be a conjugation of, in the new dictionary_key table (one row per key, indexed; migration dictionary-keys adds the table and its index, nothing else), which the loader fills beside the entries. The instance gets the text of each line looked up and keeps none of it, as the Watch screen says. Linux builds need libmpv-dev, and the app links libmpv: the snap stages Ubuntu's libmpv2, makes the update-alternatives links for BLAS and LAPACK that ffmpeg's speech recognition needs (staging never runs install scripts), and removes what the GNOME runtime and core24 already hold, as older copies of cairo, pango and librsvg hid the runtime's from its GTK. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
8.8 KiB
Halkyon Learn
Halkyon Learn is for creating, sharing and collaborating on study material. It schedules flashcards with FSRS to strengthen retention, and organises study material together with the external references it draws on. It is offline-first: everything works on the device, without a connection. And the whole stack is transparent: the app, the server and everything needed to host an instance are published, so anyone can see how an instance works and run their own.
- Offline first. Each device keeps its own database and syncs in the background
(
serverpod_offline_sync). Edits made on two devices while offline are never lost: card content is stored as append-only revisions, and a side changed differently on two devices is shown as a conflict until the person picks a version (ADR 0001, ADR 0003). - No sign-up needed. The app starts with an anonymous account; signing in with GitHub later keeps everything. A GitHub sign-in stores only the numeric GitHub ID.
- No end-to-end encryption. Content is readable by the operator of the instance, as on Mastodon; whoever trusts no operator can run their own (ADR 0002).
Targets: Linux, Windows, macOS, Android and iOS. There is no web build.
Packages
| Package | What it is | Licence |
|---|---|---|
halkyon_learn_server |
The Serverpod server: endpoints, models, migrations | SSPL-1.0 |
halkyon_learn_client |
The client, generated from the server's models | AGPL-3.0-or-later |
halkyon_learn_flutter |
The app (app.halkyon.learn) |
AGPL-3.0-or-later |
Setting up
-
Flutter v3.47.5, the version CI uses (
FLUTTER_VERSIONin.forgejo/workflows/ci.yaml). Set up each platform you build for with Flutter's own guides: Linux, Windows, Android, macOS, iOS. -
The Serverpod CLI v4.0.4:
dart install serverpod_cli@4.0.4. Pass--no-analyticsto every call; otherwise it reports the commands run to Serverpod. -
Linux, on top of Flutter's packages, the libraries of four plugins: WebKitGTK (pulled in by the sign-in package), libsecret (keeps the session), GStreamer (plays sounds) and libmpv (plays videos from Jellyfin; the app doesn't start without it). On Debian or Ubuntu, Flutter's and these together:
sudo apt-get install clang cmake ninja-build pkg-config libgtk-3-dev libstdc++-12-dev \ libwebkit2gtk-4.1-dev libsecret-1-dev libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev \ libmpv-dev -
Windows needs Developer Mode (Settings › System › For developers): building makes symbolic links, as Flutter links each plugin's native code into the build and the PDF engine's build does too, and Windows lets ordinary users make them only in Developer Mode.
-
No Docker for development:
serverpod startand the server's tests run an embedded PostgreSQL. Docker Compose is only for running an instance (below). -
For the scripts in
tool/: Python 3 (tool/app_icons.pyalso needs Pillow,python3-pil), and for releases what Releases lists.
The rich-text editor comes from our fork of AppFlowy Editor
(project-halkyon/appflowy-editor),
which flutter pub get fetches by itself.
Running it locally
The server's secrets live in halkyon_learn_server/config/passwords.yaml, which is never
committed. In a fresh clone, create it with a development and a test section, each
holding database, serverSideSessionKeyHashPepper and serviceSecret (for example from
openssl rand -base64 32). GitHub sign-in is optional: add githubClientId and
githubClientSecret of your own GitHub OAuth app to development, registered with the
callback URLs app.halkyon.learn://github-callback and http://127.0.0.1/github-callback.
Then start the server, its embedded database and the app:
cd halkyon_learn_server
serverpod --no-analytics start
serverpod start applies pending migrations and reloads on changes. The app talks to the
server named by --dart-define=SERVER_URL=…, or else to the one in
halkyon_learn_flutter/assets/config.json (http://localhost:8080). Without
--no-analytics, the Serverpod CLI reports usage to Serverpod.
Tests
tool/check.sh
runs the format check, the analyzer with the strict lints of leancode_lint, the server's
tests (against an embedded PostgreSQL, no Docker needed) and the app's tests. To run it
before every push: ln -s ../../tool/check.sh .git/hooks/pre-push. CI runs it on every
push too (.forgejo/workflows/ci.yaml).
The sync tests need a running server (serverpod start); run them in
halkyon_learn_flutter with flutter test -t requires-server.
Running an instance
deploy/ holds a Docker Compose setup: Caddy (HTTPS), the server and PostgreSQL. Copy
deploy/.env.example to deploy/.env and fill it in; it explains every value. The
licences require offering the source of what the instance runs to the people who use it.
Assistants such as Claude read people's studysets through the instance's MCP connector at
/mcp (ADR 0006): webServer in
halkyon_learn_server/config/production.yaml names its public address, as apiServer
does for the app.
learn.halkyon.eu itself deploys from main once CI is green: deploy/redeploy.sh, run by
the CI runner described in deploy/runner/, deploys only commits signed by a key in
deploy/allowed-signers. It backs up its database every hour, off-site and encrypted;
deploy/backups.md says how to set that up and how to restore.
Releases
Releases are published on code.halkyon.app,
with the APK and the snap, and the APK also in the project's F-Droid repository
(https://halkyon.app/fdroid/repo?fingerprint=E8082788756F6B7E682AA22FCEAD24B53E2A8B0912AD66D65E8C2D4F07BB94B0,
added in the F-Droid app as a repository). The Snap Store (halkyon-learn) follows.
Their builds offer learn.halkyon.eu, the project's instance, at first start; people
may choose another there, and keep accounts on several.
- Checking a download:
SHA256SUMSlists the checksum of every file, signed with the maintainer's GPG key (SHA256SUMS.asc). The APK is signed with the app's key, whose certificate has the SHA-25627:E2:43:FC:BD:C8:26:FF:98:AB:7D:BB:10:43:6C:98:C3:BE:A2:4B:2A:01:5E:94:E4:CF:C7:D1:FC:2A:00:6C(apksigner verify --print-certsshows it). - Versions follow semantic versioning:
versioninhalkyon_learn_flutter/pubspec.yaml,<major>.<minor>.<patch>+<build>. Since the last release, a commit typedfeatraises the minor number,fixorperfthe patch number, and a breaking one (feat!:or aBREAKING CHANGE:line) the major number from 1.0 on, the minor one before; other commits release nothing by themselves. Every release raises the build number: Android only installs a higher one over an older, and the server tells apps below the build it supports to update. - Making one is for maintainers, on their own machine and never in CI:
tool/release.sh, ortool/release.sh minor(ormajor,patch) to choose. It needsmainclean, pushed and green in CI, the app's key, the F-Droid repository's key and a Forgejo token; its header says more. It shows the version and its commits and asks first, then writes the version into the pubspec in a signed release commit with a signed tag, builds the APK and the snap (snap/snapcraft.yaml, built in LXD; its icon comes fromtool/app_icons.py), checks both, signs the checksums, pushes, and publishes on code.halkyon.app and in the F-Droid repository; the Snap Store once it is switched on in the script. Besides Flutter and the Android SDK, it uses snapcraft with LXD, a JDK (keytool), GnuPG, curl, jq, unzip, squashfs-tools, rsync and fdroidserver. Release builds made without the app's key are signed with the debug key and can't update a released app. - The F-Droid repository is kept by
tool/fdroid_repo.sh(fdroidserver v2.4.5, e.g.pipx install fdroidserver==2.4.5); its settings and the app's listing are intool/release/fdroid/.
More
AGENTS.md: how the code is organised and the rules it follows.docs/adr/: the architecture decisions and their reasons.CONTRIBUTING.md: licences of contributions and the sign-off.LICENSE.md: which licence covers what.