Optional AI
Local models or your chosen endpoint. Core playback works without either.
Choose a setup
| Setup | Action |
|---|---|
| On-device | Open Manage Server → AI features → Local models. Choose a catalog model, review its license and resource requirements, then download. |
| Assistant endpoint | Open Assistant setup. Enter an OpenAI-compatible endpoint, load its model catalog, choose a model, and save. |
| Speech endpoint | Configure Speech setup with an external compatible speech server, or choose a supported on-device speech model. |
What needs a model
| Feature | Needs | Without it |
|---|---|---|
| Speech subtitles | Speech model / server | Action hidden. |
| Subtitle translation | Assistant endpoint | Action hidden. |
| Episode recaps | Assistant + episode summaries | Previous episode overviews. |
| Smart collections | Assistant endpoint | Manual rule builder. |
| Semantic search | Optional embedding model | Built-in local ranking. |
| Intro / credits detection | No AI endpoint | Local detection remains available. |
Control data and resources
Features call only the configured endpoints or supervised local model processes. An external endpoint receives the text needed for the requested task; choose it accordingly. Model replies are treated as untrusted and checked against real library data.
Disable individual features in Manage Server → AI features. Downloads live in app-data/models/ and are excluded from database backups. Extra models need extra RAM.
Applies to Lumina 0.4.0. Source: operator guide