diff --git a/AGENTS.md b/AGENTS.md
index 4e62dfc..6d20b59 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -8,6 +8,10 @@ Public repo under `tinqs/live-radio`. AI agents run the station — humans liste
 - **DeepSeek** plans mood + Lyria prompts. **Google Lyria 3** renders audio.
 - Respond in English.
 
+## Taste
+
+Edit `settings.json` — DJ + chat read it every request. Default: ethnic world dubtronica.
+
 ## Session start
 
 1. Read `README.md`
diff --git a/README.md b/README.md
index f1377b9..d7768cd 100644
--- a/README.md
+++ b/README.md
@@ -14,6 +14,10 @@ Inspired by [Magenta RealTime 2](https://magenta.withgoogle.com/magenta-realtime
 | Player | FastAPI + `gateway/player.html` | Stream generated queue |
 | Live (optional) | Magenta RealTime 2 | Apple Silicon only — see below |
 
+## Taste (`settings.json`)
+
+Edit `settings.json` at the repo root — the DJ reads it on every generate and chat. Default profile: **ethnic world dubtronica** (global roots + dub space + electronic groove).
+
 ## Saved songs
 
 Every track is written to `./songs/` and **committed via Git LFS** (audio) + plain git (metadata):
diff --git a/settings.json b/settings.json
new file mode 100644
index 0000000..0ca3b1c
--- /dev/null
+++ b/settings.json
@@ -0,0 +1,53 @@
+{
+  "listener": "ozan",
+  "station": "Live Ozan Radio",
+  "taste": {
+    "summary": "Ethnic world dubtronica — global roots, dub space, and electronic groove.",
+    "genres": [
+      "ethnic world",
+      "world dub",
+      "dubtronica",
+      "dub",
+      "global electronica",
+      "desert blues",
+      "afro-dub",
+      "middle eastern dub",
+      "balkan dub",
+      "trip-hop dub"
+    ],
+    "mood": [
+      "hypnotic",
+      "warm",
+      "spacious",
+      "late-night",
+      "sunset caravan",
+      "meditative but danceable"
+    ],
+    "instruments": [
+      "sub bass",
+      "dub delay and spring reverb",
+      "hand percussion",
+      "kora or oud",
+      "nose flute or melodica",
+      "tabla",
+      "djembe",
+      "muted guitar skank",
+      "analog warmth"
+    ],
+    "tempo_bpm": [82, 108],
+    "references": "Thievery Corporation meets Sahel dub; Baaba Maal warmth over a stepper bassline; Khruangbin haze with Lee Scratch Perry space; ethnic samples woven into dubtronica, not EDM.",
+    "avoid": [
+      "big-room EDM drops",
+      "four-on-the-floor house",
+      "generic corporate lounge",
+      "overcompressed pop EDM"
+    ]
+  },
+  "dj": {
+    "variety": true,
+    "default_length": "1-2 minutes"
+  },
+  "lyria": {
+    "prefer_instrumental": true
+  }
+}
diff --git a/src/ozan_radio/dj.py b/src/ozan_radio/dj.py
index d6a50a7..7b3b1d9 100644
--- a/src/ozan_radio/dj.py
+++ b/src/ozan_radio/dj.py
@@ -8,6 +8,7 @@ import httpx
 
 from ozan_radio.config import Config
 from ozan_radio.spotify import TasteProfile
+from ozan_radio.settings import ListenerSettings, load_settings
 from ozan_radio.taste import TasteSeeds
 
 CHAT_SYSTEM = """You are the on-air DJ for Live Ozan Radio. Chat with the listener in a warm,
@@ -32,8 +33,8 @@ Your job:
 2. Pick a mood, tempo, and genre blend that feels like a natural next track.
 3. Write a Lyria prompt that produces a 1–2 minute instrumental or vocal track.
 4. Keep variety — don't repeat the same vibe twice in a row.
-5. Favor warm, groove-forward, slightly eclectic picks (Ozan's lane).
-6. West African / Sahel / desert blues / griot energy is on-brand (think Baaba Maal warmth).
+5. Follow settings.json taste profile when provided — it overrides generic defaults.
+6. Stay in the listener's lane unless they ask for something else in chat.
 
 Respond with JSON only:
 {
@@ -66,6 +67,12 @@ class DeepSeekDJ:
 
     def __init__(self, config: Config) -> None:
         self._config = config
+        self._settings = load_settings()
+
+    def _taste_block(self) -> str:
+        if self._settings:
+            return self._settings.dj_context()
+        return "No settings.json — freestyle eclectic world groove."
 
     async def _completion(self, messages: list[dict], *, json_mode: bool = False) -> str:
         self._config.require_deepseek()
@@ -98,7 +105,8 @@ class DeepSeekDJ:
         if now_playing:
             context.append(f"Currently playing: {now_playing}")
 
-        messages = [{"role": "system", "content": CHAT_SYSTEM}]
+        chat_system = f"{CHAT_SYSTEM}\n\nStation taste:\n{self._taste_block()}"
+        messages = [{"role": "system", "content": chat_system}]
         for turn in history[-8:]:
             role = "assistant" if turn["role"] == "dj" else turn["role"]
             messages.append({"role": role, "content": turn["content"]})
@@ -120,7 +128,10 @@ class DeepSeekDJ:
     ) -> TrackPlan:
         self._config.require_deepseek()
 
-        user_parts = ["Plan the next generated track for Live Ozan Radio."]
+        user_parts = [
+            "Plan the next generated track for Live Ozan Radio.",
+            f"Station taste:\n{self._taste_block()}",
+        ]
         if taste:
             user_parts.append(f"Spotify taste: {taste.summary}")
             if taste.top_genres:
diff --git a/src/ozan_radio/settings.py b/src/ozan_radio/settings.py
new file mode 100644
index 0000000..ade5606
--- /dev/null
+++ b/src/ozan_radio/settings.py
@@ -0,0 +1,64 @@
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass
+from pathlib import Path
+
+
+@dataclass
+class ListenerSettings:
+    listener: str
+    summary: str
+    genres: list[str]
+    mood: list[str]
+    instruments: list[str]
+    tempo_bpm: list[int]
+    references: str
+    avoid: list[str]
+    prefer_instrumental: bool
+
+    def dj_context(self) -> str:
+        bpm = self.tempo_bpm
+        tempo = f"{bpm[0]}-{bpm[1]} BPM" if len(bpm) >= 2 else "mid tempo"
+        lines = [
+            f"Listener taste ({self.listener}): {self.summary}",
+            f"Genres: {', '.join(self.genres)}.",
+            f"Mood: {', '.join(self.mood)}.",
+            f"Instruments / production: {', '.join(self.instruments)}.",
+            f"Tempo: {tempo}.",
+            f"References: {self.references}",
+        ]
+        if self.avoid:
+            lines.append(f"Avoid: {', '.join(self.avoid)}.")
+        if self.prefer_instrumental:
+            lines.append("Default to instrumental unless vocals are requested.")
+        return "\n".join(lines)
+
+
+def load_settings(repo_root: Path | None = None) -> ListenerSettings | None:
+    root = repo_root or Path(__file__).resolve().parents[2]
+    path = root / "settings.json"
+    if not path.exists():
+        return None
+
+    try:
+        data = json.loads(path.read_text(encoding="utf-8"))
+    except (json.JSONDecodeError, OSError):
+        return None
+
+    taste = data.get("taste", {})
+    lyria = data.get("lyria", {})
+    if not taste.get("summary") and not taste.get("genres"):
+        return None
+
+    return ListenerSettings(
+        listener=data.get("listener", "listener"),
+        summary=taste.get("summary", ""),
+        genres=taste.get("genres", []),
+        mood=taste.get("mood", []),
+        instruments=taste.get("instruments", []),
+        tempo_bpm=taste.get("tempo_bpm", []),
+        references=taste.get("references", ""),
+        avoid=taste.get("avoid", []),
+        prefer_instrumental=lyria.get("prefer_instrumental", True),
+    )
diff --git a/taste_seeds.json b/taste_seeds.json
index 37c9979..daebb5f 100644
--- a/taste_seeds.json
+++ b/taste_seeds.json
@@ -2,12 +2,12 @@
   "listener": "ozan",
   "notes": "Manual taste seeds when Spotify API is not wired. Add tracks from your library — DJ reads vibe, never replays catalog.",
   "genres": [
+    "ethnic world",
+    "dubtronica",
+    "world dub",
     "west african",
     "afro-folk",
-    "desert blues",
-    "griot",
-    "world",
-    "eclectic groove"
+    "desert blues"
   ],
   "tracks": [
     {