Teach your Navidrome server to DJ by ear

What it does

AudioMuse-AI-NV-plugin is a Navidrome extension—compiled via TinyGo to WASM and packaged as an .ndp file—that bridges Navidrome’s Instant Mix, Radio, and Artist Info features to an external AudioMuse-AI core. It intercepts similarity requests and forwards them to a dedicated backend running audio analysis workers, returning results through standard Open Subsonic APIs like getSimilarSongs2. In plain terms, it lets your media server suggest “more songs like this one” by actually analyzing the audio, not just reading tags.

The interesting bit

The plugin itself is thin glue code, but it unlocks three distinct similarity engines: a classic similar-song model, a lyric-aware “SemGrove” semantic index, and a hyperbolic projection mode. That’s unusual flexibility for a self-hosted music stack, and it exposes them through Navidrome’s standard plugin agent interface so third-party clients (Substreamer, Feishin, Symfonium, etc.) pick it up automatically.

Key highlights

• Ships as a WASM binary for Navidrome’s plugin sandbox, built with Go/TinyGo.
• Supports the Open Subsonic Sonic Similarity API extension.
• Offers three Instant Mix backends: standard audio similarity, lyrics-aware semantic search, and hyperbolic projection.
• Works across multiple tested frontends—web, iOS, Android—without client-side changes.
• Requires a separately deployed AudioMuse-AI core (Flask + workers); this repo is only the Navidrome adapter.

Caveats

• Version alignment between Navidrome, the plugin, and the AudioMuse-AI core is strict; mismatching releases can break requests.
• The ND_AGENTS order matters, and misconfiguration silently falls back to other agents rather than warning you.
• Different frontends may return different result counts because the plugin does not enforce a default limit.

Verdict

Worth a look if you run Navidrome at home and want algorithmic music discovery without sending your library to a cloud service. Skip it if you were hoping for a standalone, single-container solution; this is one piece of a larger AudioMuse-AI puzzle.