Providers · chapters · 25 of 25
DJ
Describe a playlist ("late-night 90s trip hop") and a language model picks it from your library.
facts
- status
- beta
- kind
- other
- on by default
- yes
- needs
- A library source, Apple Intelligence, a model server, or on Android a 2.6 GB model download
- platforms
- iPhone, iPad, Mac, Vision Pro, Android, Android TV, Linux, FreeBSD, Windows
Description
The DJ builds a playlist from a typed request. You write what you want to hear, for example “late-night 90s trip hop, nothing too loud”, and a language model picks albums and tracks from your own library and plays them.
The DJ only knows your library. It searches the sources you have set up for the names in your request, looks at their genre folders, and hands the model a list of what it found. The model can only choose from that list. It cannot add music you do not have, and it has not heard any of it: it works from titles, artists and genres.
Which model does the picking depends on the device:
- Apple: Apple Intelligence on the device, or a model server you name. Not on Apple TV.
- Android: a model server, or a model that Yafmp downloads and runs on the device.
- Desktop: a model server.
A model server is any server with an OpenAI-style chat API, such as Ollama, LM Studio, llama.cpp, OpenAI or OpenRouter.
Requirements
- At least one library source with music in it, such as Plex.
- One of:
- Apple: a device that runs Apple Intelligence, with Apple Intelligence turned on.
- A model server the device can reach, and an API key if the server asks for one.
- Android: 2.6 GB of free storage for the on-device model.
Setup
With Apple Intelligence there is nothing to set up. The DJ button (a sparkle) is in the toolbar.
To use a model server:
- Open Preferences → Plugins → DJ (Apple) or Settings → DJ (Android and desktop).
- Enter the server’s address in Model server, for example
http://<your server>:11434for Ollama. - Enter an API key if the server needs one.
- Choose Connect. Yafmp lists the server’s models.
- Pick a model.
To use the on-device model on Android:
- Open Settings → DJ and leave the model server empty.
- Choose Download (2.6 GB) and wait for the state to read Ready.
- Choose Say hello to test it.
Then make a playlist:
- Choose the DJ button in the toolbar.
- Type your request.
- Choose Make Playlist (Apple) or Play (Android and desktop).
Settings
| Setting | Default | What it does |
|---|---|---|
| Sources (one switch per source) | all on except Discogs and YouTube Music | Which sources the DJ searches. Fewer is faster and leaves more room for your own library. |
| Model server | empty | The server’s address. A trailing /v1 is fine. On Apple, empty means Apple Intelligence; on Android, empty means the on-device model. Labelled Model server (Ollama, LM Studio, OpenAI…) on Android and desktop. |
| API key | empty | Only for servers that need one. Labelled API key (if the server needs one) on Android and desktop. |
| Connect | — | Asks the server for its models. |
| Model | first model listed | The model that picks the playlist. |
| Download (2.6 GB) | — | Android: downloads the on-device model (Gemma 4 E2B). |
| Say hello | — | Android: asks the on-device model for one sentence and shows how long it took. |
On Apple, a line at the top says whether Apple Intelligence is available on this device. On Android and desktop, a line at the top says which model is in use.
On Apple the API key is kept in the system keychain. With iCloud Keychain on, the key, the server address and the model reach your other Apple devices.
Limits
- The DJ picks only from your library. It does not know new releases, charts or anything you have not added.
- A playlist has at most 50 tracks. An album or artist the model picks adds three of its tracks, chosen at random.
- With a model server, your request and the matching library rows are sent to that server. With Apple Intelligence or the on-device model on Android, nothing leaves the device.
- Apple Intelligence reads a small amount at a time: each search returns 15 rows in all, and the DJ gives up after two minutes. Narrow requests work better.
- On Android and desktop the model sees at most 100 candidate rows. Radio and SiriusXM are never searched.
- On Android and desktop the DJ also reads MusicBrainz tags and record labels for the artists and albums it finds. The Apple edition does not.
- The on-device model on Android takes tens of seconds to load for the first request. Its download does not resume: a dropped connection starts it again.
- The desktop has no on-device model.
- Not on Apple TV.
Troubleshooting
| Problem | Why | What to do |
|---|---|---|
| No DJ button in the toolbar (Apple) | The device has no Apple Intelligence and no model server is set | Set a model server in Preferences → Plugins → DJ, or turn on Apple Intelligence in System Settings. |
| “Set a model server in Settings → DJ” | Android and desktop: no model is ready | Set a model server, or on Android download the on-device model. |
| “pick a model in Preferences → Plugins → DJ” or “Pick a model in Settings → DJ” | A server is set but no model is chosen | Choose Connect and pick a model. |
| “the DJ didn’t find anything matching in your library” or “Nothing in the library fits that.” | The searches found nothing | Name artists or genres you have, and check that the right sources are switched on for the DJ. |
| “too much music matched for the on-device model; try a narrower request” | Apple Intelligence ran out of room | Ask for something more specific, or switch off sources the DJ does not need. |
| “the DJ gave up after two minutes; try a narrower request” | Apple Intelligence took too long | Ask for something more specific, or use a model server. |
| “The model server returned HTTP …” | The server refused the request | Check the address, the API key and that the chosen model still exists on the server. |
| “Download failed” (Android) | The connection dropped during the model download | Choose Download (2.6 GB) again. It starts from the beginning. |