Replace Spotify

What you are actually giving up#

Say it plainly: this is not a like for like replacement, and no article that tells you otherwise is being honest with you. Spotify's product is a licensing arrangement with the entire recording industry. Navidrome is a very good program for playing files you already have. Those are different things.

What goes away:

• The catalogue. Every album you have not bought. Every track a friend mentions at dinner. Every soundtrack you needed once.
• Discovery. Discover Weekly, Release Radar, autoplay radio, Blend, collaborative playlists. These are built on aggregate listening data from hundreds of millions of accounts. Nothing you run at home approximates them, and the plugin ecosystems that try are recommendation engines over a library you already chose.
• Zero-latency curiosity. The gap between "I wonder what that sounds like" and hearing it goes from three seconds to a purchase decision.
• Podcasts and exclusives, which do not exist outside Spotify at all.
• Everything downstream of the app. Spotify Connect, car head units, smart speakers, Wrapped. Some of it has a Subsonic equivalent; most of it does not.

Who this genuinely works for: people who already own a CD or FLAC collection and are paying Spotify to listen to music they have already bought. People who buy from Bandcamp habitually. People whose listening is deep and repetitive rather than broad. Households where one library serves several people. If you are not in one of those groups, the honest recommendation is to keep paying Spotify and self-host something else.

Get your data out first#

Your Spotify data is only available while the account exists, so extract it before you cancel, even if you are undecided.

1. Request both packages. On the Account Privacy page, request the Account Data package (playlists, library, streaming history, search queries, follower data, taste profile, Wrapped summaries) and the Extended Streaming History separately. The extended history is the interesting one: every item listened to across the lifetime of the account, with timestamps, platform, milliseconds played, country and the reason each track started and stopped. It arrives as JSON, by email, and it is not instant.

2. Get playlists now, with ISRCs. Waiting for a data package is fine for records; it is bad for planning. Exportify runs in your browser against Spotify's Web API, asks for read-only playlist access and writes CSVs per playlist including the track URI, album, release date, popularity and the ISRC. That last field is what lets you match a track you loved to a specific release you can buy. Track and artist names alone will not do it, because half your library is a deluxe reissue with a different title.

3. Turn the data into a shopping list. The extended streaming history plus the playlist CSVs give you a ranked list of what you actually listen to, as opposed to what you think you listen to. Sort by total milliseconds played, cut it at whatever number is honest, and buy that.

# top 100 artists by listening time, from the extended history export
jq -s 'add | group_by(.master_metadata_album_artist_name)
       | map({artist: .[0].master_metadata_album_artist_name,
              hours: (map(.ms_played) | add / 3600000 | floor)})
       | sort_by(-.hours) | .[0:100]' Streaming_History_Audio_*.json

4. Acquire, then tag, then import. Rip CDs with a verifying ripper, buy FLAC where the option exists, and run everything through MusicBrainz Picard or beets before Navidrome ever sees it. Navidrome reads tags and never writes them, so a badly tagged library looks broken in ways no server setting fixes.

Which one to pick#

Navidrome for music, without much hesitation. One static binary or one container, one SQLite file, no external database, ops load 1 out of 5, and it idles around 70 MB on a Pi 4. Mount your music read-only at /music, a data directory at /data, expose 4533, done.

What the migration actually costs#

• Disk: roughly 350 MB per album in FLAC, or 120 MB at 320 kbps MP3. A thousand albums is around 350 GB of FLAC, which is a single cheap disk. Storage is not the constraint here.
• Money: the constraint. Whatever your shopping list came to.
• Setup: under an hour. The first scan is the only slow part, roughly an hour per 100,000 tracks on spinning disks and much less on SSD, with artwork extraction dominating.
• Server load: ~70 MB idle, ~200 MB in use. Serving costs almost nothing because most clients stream the original file untouched.
• The ongoing job, which never stops: acquiring and tagging. Every new release is a purchase and a tagging session, call it ten minutes an album. That is the real recurring cost of leaving Spotify, and it is a hobby, not a chore, only if you already enjoy it.
• Backup: the navidrome.db file holds users, playlists, play counts, ratings, starred items and scrobble history, and none of it is recoverable from your audio files. Use SQLite's .backup rather than copying it live under WAL, and see Backups that actually restore. Smart playlists are .nsp JSON files in the music folder, so they travel with the library.

What breaks and how to tell early#

Subsonic clients do not use your SSO. Even with Authelia or a proxy in front of the web UI, third-party apps authenticate against Navidrome's own user database with a password. Every user needs a Navidrome password for mobile, and your proxy has to let the /rest endpoints through, or the phone simply fails to connect. Plan the proxy rules around that split rather than discovering it in a car park.

Header auth is a foot-gun. If you set ND_REVERSEPROXYUSERHEADER without setting ND_REVERSEPROXYWHITELIST to the proxy's CIDR, anything that can reach port 4533 can claim to be anyone. The first user auto-created that way becomes an admin.

Full scans can be OOM-killed. Steady-state memory is tiny, but full rescans on libraries in the hundreds of thousands of tracks have pushed resident memory past 10 GB. That is why an external scanner subprocess mode exists. If the container dies mid-scan, enable it rather than fighting the schedule.

The car, the speaker, the watch. Whatever integration you use daily is the thing to test on day one with a single album, not on day thirty with a full library.

The honest early tell. For two weeks, keep a note every time you go to play something and find you do not own it. If that number is more than a handful per month, stop. Cancelling Spotify will make you listen to less music, which is the wrong outcome from a music project.

The point of no return#

Run both for ninety days. That is longer than the other migrations on this site because the failure mode is not technical, it is that your listening quietly shrinks and you do not notice until you have unsubscribed.

Before you cancel:

1. Both Spotify data packages downloaded and stored somewhere backed up. They are unavailable after the account closes.
2. Your top 100 artists by listening time either owned or consciously abandoned.
3. The client app installed and working on every device you use, including the one in the car.
4. Ninety days of the "things I wanted and did not have" log, at fewer than five a month.
5. A tested restore of navidrome.db, because play counts and playlists are the only part of this that cannot be rebuilt from your files.

If step 4 fails, keep paying. A $12 subscription alongside a self-hosted library you actually own is a perfectly reasonable end state, and it is the one most people who try this end up at.

Next: pair it with Navidrome vs Jellyfin for music for the head to head, sort out reaching the server from a phone with Remote access without port forwarding, and see Replace Audible for the audiobook half of the same problem. The rest of the category is at Media.