sonda/README.md
tgiachi a6dbf60ac8 feat(feedback): a wishlist and "not interested"
- PUT/DELETE /wishlist/{artist} and GET /wishlist: the artists I want.
- POST/DELETE /discover/{artist}/dismiss and GET /discover/dismissed: the
  artists I am not interested in. An artist is wished or dismissed, never both,
  and both lists are mine alone.
- Discover never suggests a dismissed artist, does not use one as a seed taken
  from my listening, and marks the wished ones it shows (wishlisted).
- New artist_feedback table (migration 0012, SQLite and PostgreSQL), removed
  with the account.
- An artist I marked is not cleaned up when the list that named it is replaced,
  and goes when its last mark is taken off and nobody names it. Taking a
  provider's data back still removes what only it named, marks included.
- Spec, contract test and README updated.

Closes #35
2026-11-11 20:11:18 +01:00

308 lines
26 KiB
Markdown

# Sonda
Sonda is a privacy-first, self-hosted music server written in Go. It streams the music you already own, from your own disks, and keeps everything it knows about you on your own machine.
It is a single executable, one database (SQLite by default, PostgreSQL if you prefer) and the folders that hold your music. There is no Redis, no message broker and no cloud account.
> **Status: early.** The local-playback part works end to end: accounts, libraries, scanning, search, and streaming with HTTP Range. Suggestions of new artists work too (Discover). Radio queues, a web client and OpenSubsonic compatibility are not built yet (see [What is not there yet](#what-is-not-there-yet)). The API may still change.
## What works today
- **Accounts.** The first user to register becomes the admin and registration then closes; admins create the other users. Short-lived access tokens (JWT) plus rotating refresh tokens, so a stolen refresh token is detected.
- **Libraries.** A library is a name and a folder, managed through the API. Sonda only ever reads your music folders: it never writes, renames or deletes a file.
- **Scanner.** Reads MP3 and FLAC tags and audio properties with several workers in parallel. Rescans are incremental, renamed files keep their identity, vanished files are flagged (never deleted), and a library whose folder looks empty or unreadable (an unmounted NAS) is left alone instead of being wiped. Audio in other formats (m4a, ogg, opus, wav, wma and similar) is not read yet, but it is counted as `unsupported` in the scan result instead of disappearing silently. NAS housekeeping folders (`@Recycle`, `#recycle`, `@eaDir`, `.@__thumb`, `$RECYCLE.BIN`) are skipped.
- **Catalog.** Browse and search artists, albums and tracks, with paging, by library, artist or album. Only tracks you can actually play are listed.
- **Streaming.** `GET` and `HEAD` with `Range` support, so players can seek. Opening or prefetching a stream does not record a listen.
- **API documentation.** The full API is described in OpenAPI and browsable with Swagger UI, served by Sonda itself. The UI is embedded in the binary, so opening it contacts no other server.
- **Listening history.** Clients report explicit `play`, `complete` and `skip` events. An event counts as a listen when it is `complete`, or when it was played for at least 20 seconds (`listening.min_listen_seconds`); the rule is applied when the history is read, so changing it re-reads the past. You can read your own history, with a `counted` flag on every event. It stays in your database and is never sent anywhere.
- **Your taste.** `GET /me/taste` shows what Sonda has learned from your own listening: the artists you like and the kinds of music (tags) they have in common. A complete listen counts for more than a short one, a skip counts against the artist, and older listens weigh less (they halve every 90 days, `taste.half_life_days`; only the last ten half-lives are read). It is worked out on every request from data already in your database; nothing is sent to any provider.
- **Discover.** `GET /discover` suggests artists you may like, in your library or not, each with a score from 0 to 1 and the parts it is made of: how close it is to the artists you like (affinity), how close its tags are to yours (genre), what your listening says about it (preference) and a bonus for what you have not played (discovery). The weights are in `[scoring]` in the config. A part with no data is left out and the rest rescaled, and every answer says which seeds led to each suggestion. It starts from the artists you name with `?seed=`, or, only if you turned on `discover_from_listening`, from the artists you play most. It needs Last.fm to be switched on to ask for anything new: when Last.fm is off and a seed's similar artists are not stored yet, the answer says `disabled` (what was stored earlier is still used). See [External providers](#external-providers).
- **Wishlist and "not interested".** `PUT /wishlist/{artist}` puts an artist on your wishlist (`GET /wishlist` lists it, `DELETE` takes it off), and `POST /discover/{artist}/dismiss` hides an artist from your Discover for good (`DELETE` brings it back, `GET /discover/dismissed` lists them). An artist is wished or dismissed, never both, and both lists are yours alone. Discover marks the wished artists it shows (`wishlisted`) and never suggests a dismissed one. Nothing is downloaded or sent anywhere: the wishlist is a list for you, and for a download client you may set up later. An artist that only Last.fm named is kept while you have it marked, and goes with the rest of Last.fm's data if an admin removes it.
- **Your own settings.** `GET` and `PATCH /me/settings` hold choices that belong to you, not to the server. Everyone starts private: `discover_from_listening` is off, so this setting never sends your listening to a provider. When you turn it on, Discover may start from the artists you listen to most, and their names are sent to Last.fm (only if an admin switched it on). What Last.fm answers is kept in the shared database like any other lookup, so another local user can see, from `retrieved_at` of `GET /artists/{id}/similar`, that someone looked up an artist, but not who.
- **Tags.** `GET /artists/{id}/tags` shows what Last.fm and MusicBrainz say kind of music an artist plays, each with its own weights, and the genres of the artist's own files, read offline. See [External providers](#external-providers).
- **Similar artists.** With Last.fm switched on, `GET /artists/{id}/similar` lists artists similar to an artist, and says which of them you already have. See [External providers](#external-providers). Without it, Sonda makes no outbound network requests at all.
## What is not there yet
Radio queues, a web client, OpenSubsonic compatibility, transcoding, and the ListenBrainz integration.
## Requirements
- Go 1.27 or newer to build. The binary needs nothing else: SQLite and the tag reader are compiled in, with no CGO and no system libraries.
- `curl` and [`jq`](https://jqlang.github.io/jq/) for the demo below.
- Optional: `ffmpeg`, only to generate a few test tones for the demo.
## Build
```bash
go build -o bin/sondad ./cmd/sondad
```
## Configure
Sonda reads `./sonda.toml`, or the file given with `--config`. Every key has a default except the secret.
```bash
cp sonda.example.toml sonda.toml
# Put a random secret in it (32 characters or more). Edit the file by hand if your sed differs.
sed -i "s|^jwt_secret = .*|jwt_secret = \"$(openssl rand -hex 32)\"|" sonda.toml
```
`sonda.toml` is git-ignored because it holds the secret. Keep it readable only by the user that runs Sonda (`chmod 600 sonda.toml`).
Precedence is: built-in defaults, then the file, then environment variables. An unknown key in the file is an error, so a typo cannot be ignored silently.
| Key | Environment variable | Default | Meaning |
|---|---|---|---|
| `server.listen` | `SONDA_LISTEN` | `127.0.0.1:8080` | Address to listen on. Localhost by default. |
| `library.scan_workers` | | `4` | Files whose tags are read in parallel during a scan. |
| `database.driver` | `SONDA_DB_DRIVER` | `sqlite` | `sqlite` or `postgres`. |
| `database.path` | `SONDA_DB_PATH` | `sonda.db` | SQLite file (driver `sqlite`). |
| `database.dsn` | `SONDA_DB_DSN` | | Connection string (driver `postgres`). |
| `log.env` | `SONDA_ENV` | `production` | `production` logs JSON; `dev` logs readable text. |
| `auth.jwt_secret` | `SONDA_JWT_SECRET` | *(required)* | Signs access tokens. At least 32 characters. |
| `auth.access_ttl_minutes` | | `15` | Lifetime of an access token. |
| `auth.refresh_ttl_hours` | | `720` | Lifetime of a refresh token (30 days). |
| `listening.min_listen_seconds` | | `20` | A play counts as a listen after this long (1 to 3600), or when it is `complete`. |
| `taste.half_life_days` | | `90` | After this many days a listen weighs half as much when learning your taste (1 to 3650). |
| `scoring.weight_affinity`, `weight_genre`, `weight_preference`, `weight_discovery` | | `0.40`, `0.25`, `0.25`, `0.10` | The weights of the parts of a Discover score (0 to 10 each, not all zero). Only their proportions matter. |
Libraries and the external providers are not in the config: you set them up through the API (see the demo and [External providers](#external-providers)).
### Using PostgreSQL
```toml
[database]
driver = "postgres"
dsn = "postgres://user:password@localhost:5432/sonda?sslmode=disable"
```
Sonda creates and migrates its tables on start. The storage tests run against both databases.
## Demo
This takes you from an empty database to streaming a track. Run the server in one terminal and the commands in another.
**1. Get some music.** Point `MUSIC` at a folder of MP3 or FLAC files. The path must be absolute. If you have none, generate two 30-second test tones:
```bash
export MUSIC="$HOME/Music"
mkdir -p "$MUSIC/demo"
ffmpeg -loglevel error -y -f lavfi -i "sine=frequency=440:duration=30" -metadata title="Tone A" -metadata artist="Demo Artist" -metadata album="Demo Album" -metadata track=1 "$MUSIC/demo/a.mp3"
ffmpeg -loglevel error -y -f lavfi -i "sine=frequency=660:duration=30" -metadata title="Tone B" -metadata artist="Demo Artist" -metadata album="Demo Album" -metadata track=2 "$MUSIC/demo/b.flac"
```
**2. Start the server** (after the Build and Configure steps above):
```bash
./bin/sondad
```
**3. Check it is up, create the first user, and log in:**
```bash
BASE=http://127.0.0.1:8080/api/v1
curl -s $BASE/health
# The first user becomes the admin; registration is then closed.
curl -s -X POST $BASE/auth/register -d '{"username":"alice","password":"correct horse battery"}'
LOGIN=$(curl -s -X POST $BASE/auth/login -d '{"username":"alice","password":"correct horse battery"}')
TOKEN=$(echo "$LOGIN" | jq -r .access_token)
REFRESH=$(echo "$LOGIN" | jq -r .refresh_token)
AUTH="Authorization: Bearer $TOKEN"
```
**4. Add a library and scan it:**
```bash
curl -s -X POST $BASE/libraries -H "$AUTH" -d "{\"name\":\"Music\",\"path\":\"$MUSIC\"}" | jq
curl -s -X POST $BASE/library/scan -H "$AUTH"
until [ "$(curl -s $BASE/library/scan -H "$AUTH" | jq .running)" = false ]; do sleep 1; done
curl -s $BASE/library/scan -H "$AUTH" | jq .last
```
`found`, `added`, `unchanged`, `renamed`, `missing`, `failed` and `unsupported` tell you what the scan did. Run it again and everything is `unchanged`.
**5. Browse and search:**
```bash
curl -s "$BASE/artists?limit=5" -H "$AUTH" | jq
curl -s "$BASE/albums" -H "$AUTH" | jq '.items[] | {id, title, artist: .artist.name, tracks: .track_count}'
curl -s "$BASE/tracks?q=tone" -H "$AUTH" | jq '.items[] | {id, title, artists: [.artists[].name], format}'
```
**6. Stream a track:**
```bash
TRACK=$(curl -s "$BASE/tracks?limit=1" -H "$AUTH" | jq -r '.items[0].id')
# Ask for the first 100 bytes: the answer is 206 Partial Content with a Content-Range header.
curl -s -D - -o /dev/null -H "$AUTH" -H 'Range: bytes=0-99' $BASE/tracks/$TRACK/stream
# The whole file.
curl -s -o track.audio -H "$AUTH" $BASE/tracks/$TRACK/stream
```
A browser `<audio>` element cannot send headers, so the stream URL also accepts the access token in the query string: `$BASE/tracks/$TRACK/stream?access_token=$TOKEN`. That is the only route that does, and the query string is never logged.
**7. Report a listen.** Streaming never counts as a listen. The player reports one event per play, when the play ends or is abandoned, with the time actually played. Use one idempotency key per play: a retry with the same key is harmless, but a key that was already used is ignored even if the event differs, so an event cannot be upgraded later. An event counts as a listen when it is `complete` or lasted at least the configured minimum (20 seconds by default); a shorter play is stored too, but does not count:
```bash
curl -s -X POST $BASE/listens -H "$AUTH" -d "{\"track_id\":$TRACK,\"event_type\":\"play\",\"duration_played_ms\":25000,\"idempotency_key\":\"demo-1\"}"
curl -s -X POST $BASE/listens -H "$AUTH" -d "{\"track_id\":$TRACK,\"event_type\":\"play\",\"duration_played_ms\":3000,\"idempotency_key\":\"demo-2\"}"
# Your history, newest first. Add ?counted=true to see only what counts.
curl -s "$BASE/listens" -H "$AUTH" | jq '.items[] | {track: .track.title, event_type, duration_played_ms, counted}'
```
The answer to each report says whether it counts. A reported play time far beyond the length of the track (more than twice its length plus 10 seconds) is rejected.
**8. Refresh the session and log out:**
```bash
NEW=$(curl -s -X POST $BASE/auth/refresh -d "{\"refresh_token\":\"$REFRESH\"}")
TOKEN=$(echo "$NEW" | jq -r .access_token)
REFRESH=$(echo "$NEW" | jq -r .refresh_token)
AUTH="Authorization: Bearer $TOKEN"
curl -s -X POST $BASE/auth/logout -d "{\"refresh_token\":\"$REFRESH\"}"
```
Each refresh token works once. Presenting a spent one again, more than ten seconds after it was used, ends every session of that user.
**9. Add another user** (admins only):
```bash
TOKEN=$(curl -s -X POST $BASE/auth/login -d '{"username":"alice","password":"correct horse battery"}' | jq -r .access_token)
AUTH="Authorization: Bearer $TOKEN"
curl -s -X POST $BASE/users -H "$AUTH" -d '{"username":"bob","password":"another long password"}' | jq
```
## External providers
Sonda can ask Last.fm and MusicBrainz about artists. This is **off by default**: until an admin turns a provider on, nothing leaves your server. Today Last.fm is used for similar artists and tags, and MusicBrainz for tags.
- **Last.fm** needs an API key. Create one for your own use at `last.fm/api` (it needs a Last.fm account). No user name or password is involved and the "shared secret" is not needed. Last.fm data is for non-commercial use only, and you should credit Last.fm when you show it.
- **MusicBrainz** needs no key, but it asks every client to identify itself with a contact (a URL or an e-mail address) in the `User-Agent`. By default Sonda uses the address of the project, so no personal address is sent unless you choose to.
Set them up with `PUT /api/v1/providers/{name}` (admin), for example:
```bash
curl -s -X PUT $BASE/providers/lastfm -H "$AUTH" -d '{"api_key":"<your 32 character key>"}'
curl -s -X PUT $BASE/providers/lastfm -H "$AUTH" -d '{"enabled":true}'
curl -s $BASE/providers -H "$AUTH" | jq
curl -s -X POST $BASE/providers/lastfm/test -H "$AUTH" | jq # does the key work?
curl -s -X PUT $BASE/providers/musicbrainz -H "$AUTH" -d '{"enabled":true,"contact":"https://example.org/me"}'
```
The key is stored in the database, in clear text, and is never returned by the API or written to the log: the list only says `has_api_key`. Because the database holds it, treat database backups like secrets. Removing the key turns the provider off, and each provider describes in the list exactly what it sends.
**What leaves your server, when a provider is on:** to Last.fm, the artist name (or MusicBrainz id) being looked up, your API key, your IP address and the `User-Agent`; to MusicBrainz, only the MusicBrainz id of the artist, your IP address and a `User-Agent` that carries your contact. Never your users, your listens, track titles or file paths. Requests only happen when someone asks for them, never as a bulk pass over your library, and they are spaced out, retried with backoff, and stopped by a circuit breaker if a provider is down. **Discover** asks Last.fm for the similar artists of each seed (the seed's name or MusicBrainz id) and, for the best suggestions that have no tags yet, for the tags of that suggestion (its name); never MusicBrainz. Seeds you name are sent when you ask. Seeds taken from your listening, which are the names of the artists you play most, are sent only if you turned on `discover_from_listening` yourself; otherwise Discover refuses to run without a `seed`. The one exception to "only when someone asks" is `POST /providers/{name}/test`, which an admin sends to check a key; it sends nothing but the key.
**Similar artists.** `GET /api/v1/artists/{id}/similar` asks Last.fm for the artists similar to an artist (one of yours, or one Last.fm named earlier) and keeps the answer in your database: 7 days, or 1 day when Last.fm does not know the artist. Until then a second look costs nothing and sends nothing, and if Last.fm is down you still get the last answer, marked `stale`. If the provider is off the answer says `disabled` instead of failing. Each similar artist says whether you have it (`in_library`); the ones you do not have are stored as artists without files. They can be opened with `GET /artists/{id}` but are never part of your artist lists, and if you later add their music, the scan finds them there. Last.fm's `score` is relative to each answer (the best one is usually 1), so compare artists only inside one answer.
**Tags.** `GET /api/v1/artists/{id}/tags` asks each provider that is on, at the same time, and keeps the answers with the same rules (7 days, 1 for an unknown artist, old data if a provider fails). Each provider has its own entry with its own weights: Last.fm's are relative to the artist (1 to 100), MusicBrainz's are counts of votes, and they are never mixed. MusicBrainz also marks the tags it lists as genres. Its tags and genres are CC BY-NC-SA, so show the `attribution` that comes with them.
**Genres from your own files.** The scan also reads the genre tag of each file (several genres separated by `;`, at most five per file) and `GET /artists/{id}/tags` adds a `files` entry: the genres of the playable tracks the artist is credited on, each weighted by how many tracks have it. It needs no provider, asks nobody, and is always there.
**MusicBrainz ids come from your files, never from a guess.** If your files carry MusicBrainz artist ids in their tags (taggers like Picard write them), the scan reads them, offline, and keeps them on your artists, but only when each id can be paired with its artist for certain. MusicBrainz is asked only about artists whose id is known; for the others its entry says `no_id`. Sonda never searches MusicBrainz by name, because a search returns look-alikes and picking one would be a guess. A normal scan reads a file's tags when the file is new or changed. To pick up ids and genres for music that is already indexed, scan with `full=true` (`POST /library/scan?full=true`, or `POST /libraries/{id}/scan?full=true`): every file is read again, keeps its identity, and your listening history is untouched.
**Taking it back.** `DELETE /api/v1/providers/{name}/data` (admin) removes everything a provider gave: its similar artists, its tags and the records of when it was asked, and then the artists that only that data named and that you have no files for. Last.fm's terms ask for this when an agreement ends. Your own artists, music and listens stay, and the provider is not switched off. An artist that only that provider named goes too, even if someone had it on their wishlist or dismissed. An artist that no list names any more is also removed when the list that named it is replaced, so they do not pile up.
The artist is sent to Last.fm by its MusicBrainz id when your files gave one (if Last.fm does not know the id, by name), otherwise by its name exactly as it is written in your library, and Sonda never lets Last.fm correct it into another artist. Names and ids from Last.fm are never written onto your own artists.
## The API at a glance
Everything lives under `/api/v1`. Bodies are JSON. Errors always look like `{"error":{"code":"...","message":"..."}}`.
**Interactive documentation:** open `http://127.0.0.1:8080/api/docs/` in a browser. Use *Authorize* with the `access_token` from the login to try the protected operations. The OpenAPI file itself is at `/api/v1/openapi.yaml`. Both are public, since they describe the API and contain no data.
| Who | Method and path | What it does |
|---|---|---|
| anyone | `GET /health` | Liveness check. |
| anyone | `POST /auth/register` | Create the first user (admin). Closed afterwards. |
| anyone | `POST /auth/login`, `/auth/refresh`, `/auth/logout` | Open, renew and end a session. |
| signed in | `GET /auth/me`, `POST /auth/password` | Your account; change your password (signs you out everywhere). |
| signed in | `GET /artists`, `/albums`, `/tracks` and `/…/{id}` | Browse and search. Query: `q`, `limit` (default 50, max 200), `offset`, `library_id`, `artist_id`, `album_id`. |
| signed in | `GET /artists/{id}/tags` | What Last.fm and MusicBrainz say about the artist's kind of music, from each provider that is on. |
| signed in | `GET /artists/{id}/similar` | Artists similar to one, from Last.fm when an admin switched it on. Query: `limit` (default 20, max 50). |
| signed in | `GET`, `HEAD /tracks/{id}/stream` | Stream the original file. Also accepts `?access_token=`. |
| signed in | `POST /listens` | Record a `play`, `complete` or `skip`; the answer says whether it counts as a listen. |
| signed in | `GET /listens` | Your own history, newest first, with `counted` per event. Query: `limit`, `offset`, `counted=true`. |
| signed in | `GET /discover` | Artists you may like, with a score and its parts. Query: `seed` (an artist id, up to 5 times), `limit` (default 20, max 30), `in_library=true\|false`. |
| signed in | `GET /wishlist`, `PUT /wishlist/{id}`, `DELETE /wishlist/{id}` | Your wishlist: list it (query `limit`, `offset`), add an artist, take one off. |
| signed in | `GET /discover/dismissed`, `POST /discover/{id}/dismiss`, `DELETE /discover/{id}/dismiss` | The artists you are not interested in: list them, dismiss one, bring it back. |
| signed in | `GET /me/taste` | The artists and tags Sonda learned you like, from your own listening. Query: `limit` (default 20, max 50). |
| signed in | `GET /me/settings`, `PATCH /me/settings` | Your own settings (today: `discover_from_listening`, off by default). |
| signed in | `GET /library/scan` | Scan status and the last result. |
| admin | `GET`, `POST /libraries`; `PATCH`, `DELETE /libraries/{id}` | Manage libraries. Deleting one removes its index entries, never your files. |
| admin | `POST /library/scan`, `POST /libraries/{id}/scan` | Scan everything, or one library. `?full=true` reads every file again. |
| admin | `GET /providers`, `PUT /providers/{name}`, `POST /providers/{name}/test`, `DELETE /providers/{name}/data` | See, set up and check Last.fm and MusicBrainz, and remove what they gave. The API key is write-only. |
| admin | `GET`, `POST /users`; `PATCH`, `DELETE /users/{id}`; `PUT /users/{id}/password` | Manage users and the admin role. The last admin cannot be demoted or deleted. |
A library path must be an absolute, existing folder, never the filesystem root, and never inside or around another library. Search is a case-insensitive substring match: `BJÖRK` finds `Björk`, but accents are kept, so `bjork` does not.
## Security notes
- Sonda listens on localhost only, unless you change `server.listen`. To use it from other machines, put it behind a reverse proxy that terminates HTTPS.
- Passwords are hashed with Argon2id. Refresh tokens are stored only as hashes. An access token is checked against the database on every request, so it stops working as soon as its account is deleted, even if the same user id is given to a new account.
- Music folders are read-only to Sonda. Streaming opens files through `os.Root`, so a path can never leave its library, even if a symlink is swapped in later.
- File paths are never returned by the catalog, and audio file paths are not logged.
- Provider API keys are write-only through the API and never logged. They sit in the database, so protect your backups.
## Back up
Back up the database and `sonda.toml`. Your music stays wherever you keep it.
- **SQLite:** `sqlite3 sonda.db ".backup sonda-backup.db"` works while Sonda runs. Or stop Sonda and copy `sonda.db` (and `sonda.db-wal` if it exists).
- **PostgreSQL:** `pg_dump` the database.
- `sonda.toml` holds the secret that signs tokens; treat the copy like a password.
## Development
```bash
go vet ./...
go test -race ./...
```
The default test run needs no network and uses SQLite only. To run the same storage tests against PostgreSQL as well, start a throwaway server and point the tests at it:
```bash
docker run -d --rm --name sonda-test-pg -e POSTGRES_PASSWORD=sonda -e POSTGRES_DB=sonda -p 127.0.0.1:55432:5432 postgres:17-alpine
SONDA_TEST_POSTGRES_DSN='postgres://postgres:sonda@127.0.0.1:55432/sonda?sslmode=disable' go test -race ./...
docker rm -f sonda-test-pg
```
Tests that read real audio need `ffmpeg` and are skipped when it is missing.
Layout:
```
cmd/sondad/ the server: wiring, config, logging, shutdown
internal/api/http/ the REST API (thin handlers) and the docs routes
internal/api/openapi/ the OpenAPI spec, embedded in the binary
internal/auth/ users, roles, passwords, access and refresh tokens
internal/catalog/ browsing and searching artists, albums, tracks
internal/library/ libraries, the scanner, the event hub, the tag reader
internal/discovery/ similar artists: what the providers said, kept and refreshed
internal/providers/ external providers: settings, the outbound HTTP client, the Last.fm and MusicBrainz clients
internal/listening/ listen events
internal/taste/ what the listener likes, learned from the listening history
internal/discover/ the score of an artist you may like, and the suggestions
internal/feedback/ the wishlist and the artists a user is not interested in
internal/usersettings/ the settings that belong to one user
internal/storage/ SQLite and PostgreSQL through bun, the repositories
internal/streaming/ opening a track's file safely
migrations/ SQL migrations, one folder per database
```
The API description is `internal/api/openapi/openapi.yaml`, written by hand and embedded in the binary. Two tests keep it in step with the code: one fails when a route and the spec disagree in either direction, and a contract test sends real requests through every operation and validates each response (status, headers, JSON body) against the spec. Adding or changing an endpoint therefore means updating the spec in the same change. Response objects are closed (`additionalProperties: false`), so a field added in the code and forgotten in the spec fails the build.
Migrations are plain SQL files in `migrations/sqlite` and `migrations/postgres`. They must describe the same schema, and a test checks that both folders hold the same files. Add a new numbered file to both whenever the schema changes.
## Third-party code
Audio tags are read with `go.senan.xyz/taglib`, which embeds [TagLib](https://github.com/taglib/taglib) compiled to WebAssembly and is licensed under the LGPL 2.1. The Swagger UI assets come from `github.com/swaggest/swgui` (Apache-2.0), which embeds [Swagger UI](https://github.com/swagger-api/swagger-ui) (Apache-2.0). The other dependencies are listed in `go.mod`.