AI radio stations that make it up as they go.
Slopify runs 24/7 radio stations inside Discord voice channels. An AI DJ writes the show in real time — the songs, the ads, the news, the callers — and every second of audio is generated as it airs. Nothing is prerecorded. Nothing exists until you hear it.
🎮 For gaming crews
Drop a station into your voice channel. The DJ knows who's in the room, greets people who join, takes live call-ins, plays requests, and learns what your server loves — one 🔥 at a time.
Add to Discord🧑💻 For vibe coders in the works
Create your own station with an API key: your host, your vibe, your keys. Jam to it from a TUI while you ship, and publish its most cursed bangers to a public market. The draft spec is published — none of it is live yet.
Read the draft Station API specHow it works
Three steps from silence to a live broadcast. No config files, no dashboard.
One click. It needs voice + message permissions in your server — that's the whole ask.
An admin runs /setup and pastes two API keys:
ElevenLabs (music + voices, paid credits) and
OpenRouter (the DJ's brain — free models work, so it can cost $0).
Keys are validated live, stored encrypted, and only ever used for your server's broadcast.
Join a voice channel and run /tunein. The DJ takes it from there.
| Command | What it does |
|---|---|
/tunein | Start broadcasting into your current voice channel |
/callin message | Call the station live — the DJ answers, airs your words in a caller's voice, and reacts |
/steer direction | Steer the show live ("more synthwave", "less ads") — obeyed for the rest of the session |
/skip | Yank the current segment off the air |
/status | Buffer + queue health |
/setup | Store or update your server's API keys (add clear:true to wipe them) |
/tuneout | Off the air |
Under the hood
The interesting engineering, honestly stated.
The DJ runs on OpenRouter free models — it discovers the live free pool at runtime, benches models that 404/429, and retries down the chain. A dead model never silences a station.
Songs come from ElevenLabs Music v2: mid-song genre morphs, sound effects baked into tracks, fast rap. Renders land in seconds.
Once per session, after enough call-ins and reactions, the DJ writes a song with real lyrics about the actual people in the room — via a composition plan with explicit verse/chorus lyrics.
Generation runs ahead of playout into a buffered queue; call-ins interrupt and jump it. TTS goes through an ffmpeg broadcast chain (compression/EQ) into Discord as Opus.
🔥 / 💀 / 👎 reactions and skips feed a per-server taste memory in Postgres. Hot bits escalate, tired bits retire, requested genres stick.
Every server runs on its own keys, encrypted at rest, hot-swappable mid-broadcast. Wipe them any time.
Powered by
The three services doing the impossible parts. Slop responsibly.
OpenRouter
The brain's switchboard. Every DJ line routes across the live free-model pool —
discovered at runtime, rebalanced when models rotate out.
Hermes · Nous Research
Hermes 3 405B is a workhorse of the brain chain — an open-weights model with
exactly the right amount of unhinged for radio.
ElevenLabs
Every song and every voice. Music v2 renders full tracks — genre morphs,
embedded SFX, lyrics about your room — in seconds, at air time.
Bring your own agent
Everything a coding agent needs to work with Slopify — the command surface, the architecture, and how to spin up a station of your own — lives in one markdown file:
curl -s https://slopify.live/agents.md
Paste it into Claude Code, Cursor, or whatever you run.
It covers adding the hosted bot, tuning your server's persona, and self-hosting an entire
network. Also served at /llms.txt.
On the dial
The network is clean. The stations are not. Every station gets its own host, its own lore, and its own completely unreasonable vibe. Browse the full directory →