bvandeusenandClaude Opus 5
14aa22198f
The Discover request surface was the one recommendation surface still on its M5c implementation from early May. #796's taste profile, #1488's taste_unheard bucket and #1490's folksonomy enrichment all modernized in-library surfaces; this one was never in scope for any of them, so it still projected raw likes + plays through artist_similarity_unmatched. Two defects fall out of that signal, `5*liked + Σexp(-age/halflife)` summed over every play of the artist. It is unbounded, and contribution is signal × similarity — so a handful of heavily-played artists monopolize all twelve slots, and their share GROWS the more the user listens. The surface entrenched harder the better it knew you, which is exactly backwards and matches the reported "goes stale once it has a strong signal of your taste". It also counted every play_event with no was_skipped filter, so skipping an artist repeatedly INCREASED its signal and pushed more of its neighbours at the user. ListMostPlayedTracksForUser and the taste engine both filter skips; this query was the odd one out. Seeds now come from taste_profile_artists.weight, which the taste engine has already engagement-graded, time-decayed and signed — an artist the user drifted away from stops contributing instead of accumulating forever, and can even contribute negatively. Tiered per rule #131 rather than hard-switched: tier 1 is the profile, tier 2 is likes + completed plays for a user who has no profile rows yet (new account, or before the first daily recompute), so the surface never empties. The old unfiltered-play signal is gone, not kept behind a toggle. The signal is also log-damped, so one artist cannot take every slot even when its weight dwarfs the rest. $2 stays wired to the tier-2 decay: it is genuinely still used there, and dropping the parameter would have changed the generated signature. sqlc's image is not on this workstation and the change preserves the query signature exactly — same three params, same seven columns — so only the embedded SQL const moves. Both copies are edited and verified byte-identical rather than pulling a container onto the operator's machine; a malformed query fails the integration lane loudly, which is the real check either way. Four integration tests cover what changed: a taste weight alone seeds with no like or play; tier 2 does not run alongside tier 1; a non-positive weight never seeds (guarded by a second positive row, so an empty tier 1 can't make it pass for the wrong reason); and skip-only history seeds nothing. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Minstrel
A self-hosted music server that thinks for you. Smart shuffle, contextual likes, ListenBrainz-aware radio, and Lidarr automation — server-side, so every client (web, mobile, Subsonic third-party) gets the same intelligence.
State and intelligence belong on the server, not the client.
Highlights
- OpenSubsonic-compatible. Existing Subsonic clients (DSub, Symfonium, play:Sub, etc.) connect with no special configuration.
- Server-side smart shuffle. Track-similarity vectors, dual-like model (general + contextual), and session memory keep mixes coherent across devices.
- ListenBrainz radio. Session-aware "more like this" pulls from ListenBrainz similarity data, not a static genre tag.
- Lidarr integration. Triggered scans, request-driven album imports, and a quarantine flow when something doesn't fit.
- Built-in web SPA. Full-feature library, search, queue, playlists, and admin — no separate frontend container to deploy.
- Native Android client, shipped with the server. The signed APK is bundled into every image and attached to each release — sideload it once, then the app self-updates straight from your own server (no app store, no separate download to track).
Quickstart
# compose.yaml
services:
minstrel:
image: git.fabledsword.com/bvandeusen/minstrel:latest
ports: ['4533:4533']
volumes:
# Your music library. Point ./music at wherever your audio files
# live. Mounted read-only — Minstrel never writes to your library.
- ./music:/music:ro
# Generated data: playlist cover collages, artist art, caches.
# The path must match MINSTREL_STORAGE_DATA_DIR, which the image
# sets to /app/data — keep this mount on /app/data or your cache
# won't survive a container recreate.
- minstrel-data:/app/data
environment:
MINSTREL_DATABASE_URL: postgres://minstrel:minstrel@db:5432/minstrel?sslmode=disable
# Colon-separated library roots to scan; must match the container
# path of the read-only music mount above (/music here).
MINSTREL_LIBRARY_SCAN_PATHS: /music
depends_on: [db]
db:
image: postgres:17
environment:
POSTGRES_USER: minstrel
POSTGRES_PASSWORD: minstrel
POSTGRES_DB: minstrel
# Postgres data dir — users, likes, play history, sessions, settings.
# The one volume you must never lose; back it up with pg_dump.
volumes: [pgdata:/var/lib/postgresql/data]
volumes:
minstrel-data:
pgdata:
docker compose up -d
First run
With the stack up, a handful of in-app steps get you to a working library. Use your own host in place of localhost if you're reaching the server over a LAN/VPN address (plain http:// is fine — no TLS required).
1. Create your admin account. Visit http://localhost:4533/register. The first account on a fresh instance is automatically the administrator; later users join through the same form or an invite token (step 5).
2. Let the first library scan finish. scan_on_startup is on by default, so Minstrel walks your mounted library on boot and imports artists, albums, and tracks — no button to press. Watch progress (and re-scan any time) on the Admin page (/admin); the scan runs in stages and is incremental, so later restarts only pick up what changed.
3. (Optional) Name the instance and wire up integrations. In admin Settings → Integrations (/admin/integrations), add a ListenBrainz token (scrobbling + similarity radio) and/or a Lidarr URL + API key (the request flow). These live in the UI and apply without a restart; the display name can also be set via MINSTREL_BRANDING_APP_NAME.
4. Install the Android app. Open Settings (/settings) and use the Install the Android app card to download the APK that ships inside this server image, then sign in with the same account. From then on the app self-updates straight from your server.
5. Invite the rest of the household. From admin Users (/admin/users), generate an invite token (or enable open registration). Each person gets their own account, so likes, play history, and recommendations stay per-user.
For the full configuration surface, see config.example.yaml.
Configuration
Most operators only need the env vars in the quickstart above. A few extras worth knowing:
MINSTREL_BRANDING_APP_NAME— rename the instance ("Family Jukebox", "Office Music"). Surfaces in the header, browser tab, and OG share previews.MINSTREL_STORAGE_DATA_DIR— where generated artefacts (playlist cover collages, artist art, caches) are written. The container image sets this to/app/data, which is why the quickstart mounts theminstrel-datavolume there.MINSTREL_LIBRARY_SCAN_PATHS— colon-separated list of music library roots to scan. Supports multiple roots (/music:/podcasts).
ListenBrainz integration (per-user scrobble + similarity tokens) and Lidarr integration (URL + API key) are configured through the admin Settings UI rather than env vars or yaml — per Minstrel's "config in UI" rule, integration settings live where operators can edit them without restarting.
Most operational keys have a MINSTREL_<SECTION>_<FIELD> env override. Recommendation and events tuning are yaml-only. See config.example.yaml for the authoritative surface.
Updating
Image tags (git.fabledsword.com/bvandeusen/minstrel:<tag>):
:latest— the newest blessed image. Moves on everymainpush and every release. Recommended for most operators.:vYYYY.MM.DD— immutable per-day release tags. Pin one of these for a deployment you don't want moving under you. (Per-day CalVer — no trailing patch digit; a same-day re-cut moves the tag forward.):main— the rolling post-merge tip. Same image as:latestat push time; choose it if you want to trackmainexplicitly rather than the release line.
Every :latest and every :vYYYY.MM.DD bundles the current signed Android APK, so the in-app update channel is always live. Database migrations run automatically at startup; rollbacks require restoring a Postgres dump.
Specs
Authoritative scope lives under docs/:
- Server spec — current implementation focus.
- Client spec — Flutter companion app.
Development
Two concurrent dev processes:
- Backend:
docker compose up— Postgres + Minstrel on:4533. - Frontend:
cd web && npm install && npm run dev— Vite dev server on:5173with HMR. The Vite server proxies/api/*and/rest/*to:4533so session cookies work.
Testing
- Unit + race (no DB):
make test-short. - Full suite incl. integration tests:
make test-integration. This runs against a dedicatedminstrel_testdatabase so a test run never truncates your devminstreldata (admin user, library, likes). It brings up the compose Postgres and creates the test DB if missing. - CI runs both: a fast
go test -short -racegate plus an integration job with its own ephemeral Postgres (.gitea/workflows/test-go.yml).
Production build
docker build -t minstrel . runs the SvelteKit build inside a node stage, copies the output into the golang stage, and //go:embeds it into the final binary. The container serves the SPA from / alongside the API surfaces; no separate static-file server is required.
Branches
- Day-to-day work happens on
dev(or feature branches merged intodev). mainis protected — changes land via PR fromdev.- Releases are cut by tagging
v*offmain; the release workflow builds and pushes the container image to the Gitea registry.
Task and milestone tracking: Fable (Minstrel project, id 12).
License
See LICENSE.