fix: don't expect/use local config for remote servers

1 files changed, 110 insertions, 43 deletions

diff --git a/app/ui/hub.py b/app/ui/hub.py
index ef396e5..b603dd0 100644
--- a/app/ui/hub.py
+++ b/app/ui/hub.py
@@ -232,25 +232,50 @@ def _convert_menu(stdscr, statuses) -> Optional[tuple]:
 
 
 def _convert_audiocpp(stdscr, statuses) -> Optional[tuple]:
-    """Collect audio.cpp run settings by reading app/audio.cpp/server.json."""
+    """Collect audio.cpp run settings for a managed or remote server.
+
+    With a local checkout configured (its server.json), the menus are fed
+    from that file — the config of the server this tool manages. Without
+    one, the running server is external and nothing is known about it
+    locally, so its model and voice lists are queried live instead (the
+    same GET /v1/models and GET /v1/audio/voices endpoints the converter
+    resolves at run time).
+    """
     checkout = audiocpp_backend.find_local_checkout()
     server_json = checkout / "server.json" if checkout else None
-    if not server_json or not server_json.exists():
-        tui.flash(stdscr, "No server.json found in the audio.cpp checkout. "
-                  "Run 'Set up a backend' first.")
-        return None
-    try:
-        data = json.loads(server_json.read_text(encoding="utf-8"))
-    except (OSError, ValueError):
-        tui.flash(stdscr, f"Could not read {server_json}.")
-        return None
-    models = data.get("models") or []
-    if not models:
-        tui.flash(stdscr, "No model entries in server.json. Reconfigure "
-                  "audio.cpp first.")
-        return None
-    model_options = [(f"{m.get('id')}  ({m.get('family')}, {m.get('task', 'tts')})",
-                      m.get("id")) for m in models]
+    local = bool(server_json and server_json.exists())
+    url = config.AUDIOCPP_API_URL
+
+    if local:
+        try:
+            data = json.loads(server_json.read_text(encoding="utf-8"))
+        except (OSError, ValueError):
+            tui.flash(stdscr, f"Could not read {server_json}.")
+            return None
+        models = data.get("models") or []
+        if not models:
+            tui.flash(stdscr, "No model entries in server.json. Reconfigure "
+                      "audio.cpp first.")
+            return None
+    else:
+        # Remote flow: the backend only reaches the convert menu while a
+        # server is running, so query it — the local config says nothing
+        # about an external server.
+        models = audiocpp_backend.fetch_server_models(url)
+        if models is None:
+            tui.flash(stdscr, f"Could not list models from the audio.cpp "
+                      f"server at {url}. Is an audiocpp_server answering "
+                      "there?")
+            return None
+        if not models:
+            tui.flash(stdscr, f"The audio.cpp server at {url} hosts no "
+                      "model entries.")
+            return None
+        data = {}
+
+    model_options = [(f"{m.get('id')}  ({m.get('family') or '?'}, "
+                      f"{m.get('task') or 'tts'})", m.get("id"))
+                     for m in models]
     model_id = tui.menu(stdscr, "Select the audio.cpp model to use",
                         model_options, back_value=_GO_BACK)
     if model_id is _GO_BACK or model_id is None:
@@ -258,16 +283,30 @@ def _convert_audiocpp(stdscr, statuses) -> Optional[tuple]:
     entry = next((m for m in models if m.get("id") == model_id), {})
     family = entry.get("family")
     task = entry.get("task", "tts")
+    if not local:
+        # Servers predating the family/task fields omit them; mirror the
+        # converter's defaults (_resolve_family/_resolve_task): unknown
+        # family means qwen3_tts, a missing task plain tts.
+        family = family or AUDIOCPP_FAMILY_QWEN3_TTS
+        task = task or "tts"
+
+    def _voices() -> Optional[list]:
+        """Voice names for MODEL_ID, or None when a remote server's voice
+        list cannot be queried. Locally: the voice_dir's .wav stems;
+        remotely: GET /v1/audio/voices."""
+        if local:
+            voice_dir = data.get("voice_dir")
+            return _list_voices(voice_dir) if voice_dir else []
+        return audiocpp_backend.fetch_server_voices(url, model_id)
 
     # Voice: optional for qwen3_tts (built-in speaker), required otherwise.
     voice = None
-    voice_dir = data.get("voice_dir")
-    voices = _list_voices(voice_dir) if voice_dir else []
     if task == "vdes":
         # Voice design: no voice, instructions required.
         pass
     elif family == AUDIOCPP_FAMILY_QWEN3_TTS:
         # Speaker mode available; voice optional.
+        voices = _voices() or []
         if voices:
             opts = [("(built-in speaker)", None)] + [(v, v) for v in voices]
             voice = tui.menu(stdscr, "Voice", opts, back_value=_GO_BACK)
@@ -276,10 +315,21 @@ def _convert_audiocpp(stdscr, statuses) -> Optional[tuple]:
         else:
             voice = None
     else:
+        voices = _voices()
+        if voices is None:
+            tui.flash(stdscr, f"Could not list voices from the audio.cpp "
+                      f"server at {url}.")
+            return None
         if not voices:
-            tui.flash(stdscr, f"This model needs a --voice but voice_dir "
-                      f"{voice_dir} has no .wav voices. Reconfigure "
-                      "audio.cpp or add voices.")
+            if local:
+                tui.flash(stdscr, f"This model needs a --voice but voice_dir "
+                          f"{data.get('voice_dir')} has no .wav voices. "
+                          "Reconfigure audio.cpp or add voices.")
+            else:
+                tui.flash(stdscr, f"This model needs a --voice but the "
+                          f"server at {url} lists none for '{model_id}'. "
+                          "Configure voice presets or a voice_dir on the "
+                          "server.")
             return None
         voice = tui.menu(stdscr, "Select the voice to clone", [(v, v) for v in voices],
                          back_value=_GO_BACK)
@@ -344,29 +394,46 @@ def _convert_qwen(stdscr) -> Optional[tuple]:
 
 
 def _convert_faster(stdscr) -> Optional[tuple]:
-    """Collect faster run settings: pick a voice from voices.json."""
+    """Collect faster run settings: pick or type a voice name.
+
+    With a local checkout's voices.json the picker lists it (the config of
+    the server this tool manages). Without one, the running server was
+    configured elsewhere and its voice names are unknown here, so the name
+    is typed instead — safe for any value, since the server falls back to
+    its first configured voice when the name is not defined.
+    """
     checkout = faster_backend._checkout()
     voices_json = checkout / "voices.json"
-    if not voices_json.exists():
-        tui.flash(stdscr, f"No voices.json at {voices_json}. Run 'Set up a "
-                  "backend' for faster first.")
-        return None
-    try:
-        voices = json.loads(voices_json.read_text(encoding="utf-8"))
-    except (OSError, ValueError):
-        tui.flash(stdscr, f"Could not read {voices_json}.")
-        return None
-    if not voices:
-        tui.flash(stdscr, "voices.json has no voices. Reconfigure faster.")
-        return None
-    default = config.FASTER_VOICE if config.FASTER_VOICE in voices else \
-        next(iter(voices))
-    voice = tui.menu(
-        stdscr, "Select the voice to clone",
-        [(k, k) for k in voices],
-        default_index=list(voices).index(default), back_value=_GO_BACK)
-    if voice is _GO_BACK or voice is None:
-        return None
+    voices = None
+    if voices_json.exists():
+        try:
+            voices = json.loads(voices_json.read_text(encoding="utf-8"))
+        except (OSError, ValueError):
+            tui.flash(stdscr, f"Could not read {voices_json}.")
+            return None
+        if not voices:
+            tui.flash(stdscr, "voices.json has no voices. Reconfigure faster.")
+            return None
+    if voices is None:
+        # No local voices.json: prompt for a server-side voice name.
+        voice_text = tui.line_edit(
+            stdscr, "Server-side voice to clone with "
+            "(blank uses the server's first voice)",
+            config.FASTER_VOICE,
+            validate=lambda s: None if s.strip() else "Enter a voice name",
+            back_value=_GO_BACK)
+        if voice_text is _GO_BACK:
+            return None
+        voice = voice_text.strip()
+    else:
+        default = config.FASTER_VOICE if config.FASTER_VOICE in voices else \
+            next(iter(voices))
+        voice = tui.menu(
+            stdscr, "Select the voice to clone",
+            [(k, k) for k in voices],
+            default_index=list(voices).index(default), back_value=_GO_BACK)
+        if voice is _GO_BACK or voice is None:
+            return None
     common_kw = _common_options(stdscr)
     if common_kw is None:
         return None