Spotlight
The DJ runs on whatever brain you give it
SUB/WAVE talks to a local Ollama model out of the box, with no API key. Want a cloud model instead? Switch provider in admin and every call reroutes, no redeploy needed.
The DJ's banter, its track picks, the way it reads the room: all of that comes from a language model. SUB/WAVE doesn't lock you to one. You choose which model thinks for your station, and you can change your mind whenever.
What you get by default
Fresh installs talk to a local Ollama box. It needs no API key and nothing leaves your network, so the station works fully offline and fully private from the first boot. A homelab model is slower than a cloud one, but it's reliable and it's yours.
How to switch
Open admin, go to the LLM settings, and pick a provider:
- Ollama (local, the default)
- Any OpenAI-compatible server (llama.cpp, vLLM, LM Studio) by pasting its base URL
- OpenAI, Anthropic, Google, DeepSeek, OpenRouter, or the Vercel AI Gateway
Drop in a model name and an API key if the provider needs one, then save. That's it. (Running GGUF models through locca? It gets first-class treatment of its own.) Every call the DJ makes reroutes to the new model immediately, whether that's picking the next track, writing an intro, or matching a request. No rebuild, no restart, no code change.
Why it matters
You get to tune the trade-off yourself. Run a small local model and keep everything on your own hardware. Or point it at a sharper cloud model when you want quicker, wittier links between songs. Same station, different brain, swapped in under a minute.