fix(telegram): separate voice transcription client (#166)

docs/how-to/voice-notes.md

@@ -9,6 +9,10 @@ Enable transcription so voice notes become normal text runs.
     ```sh
     takopi config set transports.telegram.voice_transcription true
     takopi config set transports.telegram.voice_transcription_model "gpt-4o-mini-transcribe"
+
+    # local OpenAI-compatible transcription server (optional)
+    takopi config set transports.telegram.voice_transcription_base_url "http://localhost:8000/v1"
+    takopi config set transports.telegram.voice_transcription_api_key "local"
     ```
 
 === "toml"
@@ -17,13 +21,17 @@ Enable transcription so voice notes become normal text runs.
     [transports.telegram]
     voice_transcription = true
     voice_transcription_model = "gpt-4o-mini-transcribe" # optional
+    voice_transcription_base_url = "http://localhost:8000/v1" # optional
+    voice_transcription_api_key = "local" # optional
     ```
 
-Set `OPENAI_API_KEY` in your environment (uses OpenAI’s transcription API).
+Set `OPENAI_API_KEY` in your environment (or `voice_transcription_api_key` in config).
 
-To use a local OpenAI-compatible Whisper server, also set `OPENAI_BASE_URL`
+To use a local OpenAI-compatible Whisper server, set `voice_transcription_base_url`
-(for example, `http://localhost:8000/v1`) and a dummy `OPENAI_API_KEY` if your server ignores it.
+(and `voice_transcription_api_key` if the server expects one). This keeps engine
-If your server requires a specific model name, set `voice_transcription_model` (for example, `whisper-1`).
+requests on their own base URL without relying on `OPENAI_BASE_URL`. If your server
+requires a specific model name, set `voice_transcription_model` (for example,
+`whisper-1`).
 
 ## Behavior
 

docs/reference/config.md

@@ -51,6 +51,8 @@ If you expect to edit config while Takopi is running, set:
 | `voice_transcription` | bool | `false` | Enable voice note transcription. |
 | `voice_max_bytes` | int | `10485760` | Max voice note size (bytes). |
 | `voice_transcription_model` | string | `"gpt-4o-mini-transcribe"` | OpenAI transcription model name. |
+| `voice_transcription_base_url` | string\|null | `null` | Override base URL for voice transcription only. |
+| `voice_transcription_api_key` | string\|null | `null` | Override API key for voice transcription only. |
 | `session_mode` | `"stateless"`\|`"chat"` | `"stateless"` | Auto-resume mode. Onboarding sets `"chat"` for assistant/workspace. |
 | `show_resume_line` | bool | `true` | Show resume line in message footer. Onboarding sets `false` for assistant/workspace. |
 

docs/reference/transports/telegram.md

@@ -33,6 +33,10 @@ Configuration (under `[transports.telegram]`):
     ```sh
     takopi config set transports.telegram.voice_transcription true
     takopi config set transports.telegram.voice_transcription_model "gpt-4o-mini-transcribe"
+
+    # local OpenAI-compatible transcription server (optional)
+    takopi config set transports.telegram.voice_transcription_base_url "http://localhost:8000/v1"
+    takopi config set transports.telegram.voice_transcription_api_key "local"
     ```
 
 === "toml"
@@ -40,16 +44,19 @@ Configuration (under `[transports.telegram]`):
     ```toml
     voice_transcription = true
     voice_transcription_model = "gpt-4o-mini-transcribe" # optional
+    voice_transcription_base_url = "http://localhost:8000/v1" # optional
+    voice_transcription_api_key = "local" # optional
     ```
 
-Set `OPENAI_API_KEY` in the environment. If transcription is enabled but the API key
+Set `OPENAI_API_KEY` in the environment (or `voice_transcription_api_key` in config).
-is missing or the audio download fails, takopi replies with a short error and skips
+If transcription is enabled but no API key is available or the audio download fails,
-the run.
+takopi replies with a short error and skips the run.
 
-To use a local OpenAI-compatible Whisper server, also set `OPENAI_BASE_URL` (for
+To use a local OpenAI-compatible Whisper server, set `voice_transcription_base_url`
-example, `http://localhost:8000/v1`) and a dummy `OPENAI_API_KEY` if your server
+(and `voice_transcription_api_key` if the server expects one). This keeps engine
-ignores it. If your server requires a specific model name, set
+requests on their own base URL without relying on `OPENAI_BASE_URL`. If your server
-`voice_transcription_model` (for example, `whisper-1`).
+requires a specific model name, set `voice_transcription_model` (for example,
+`whisper-1`).
 
 ### Trigger mode (mentions-only)
 

src/takopi/cli/doctor.py

@@ -47,9 +47,14 @@ def _doctor_file_checks(settings: TakopiSettings) -> list[DoctorCheck]:
 def _doctor_voice_checks(settings: TakopiSettings) -> list[DoctorCheck]:
     if not settings.transports.telegram.voice_transcription:
         return [DoctorCheck("voice transcription", "ok", "disabled")]
+    api_key = settings.transports.telegram.voice_transcription_api_key
+    if api_key:
+        return [
+            DoctorCheck("voice transcription", "ok", "voice_transcription_api_key set")
+        ]
     if os.environ.get("OPENAI_API_KEY"):
         return [DoctorCheck("voice transcription", "ok", "OPENAI_API_KEY set")]
-    return [DoctorCheck("voice transcription", "error", "OPENAI_API_KEY not set")]
+    return [DoctorCheck("voice transcription", "error", "API key not set")]
 
 
 async def _doctor_telegram_checks(

src/takopi/settings.py

@@ -98,6 +98,8 @@ class TelegramTransportSettings(BaseModel):
     voice_transcription: bool = False
     voice_max_bytes: StrictInt = 10 * 1024 * 1024
     voice_transcription_model: NonEmptyStr = "gpt-4o-mini-transcribe"
+    voice_transcription_base_url: NonEmptyStr | None = None
+    voice_transcription_api_key: NonEmptyStr | None = None
     session_mode: Literal["stateless", "chat"] = "stateless"
     show_resume_line: bool = True
     forward_coalesce_s: float = Field(default=1.0, ge=0)

src/takopi/telegram/backend.py

@@ -138,6 +138,8 @@ class TelegramBackend(TransportBackend):
             voice_transcription=settings.voice_transcription,
             voice_max_bytes=int(settings.voice_max_bytes),
             voice_transcription_model=settings.voice_transcription_model,
+            voice_transcription_base_url=settings.voice_transcription_base_url,
+            voice_transcription_api_key=settings.voice_transcription_api_key,
             forward_coalesce_s=settings.forward_coalesce_s,
             media_group_debounce_s=settings.media_group_debounce_s,
             topics=settings.topics,

src/takopi/telegram/bridge.py

@@ -124,6 +124,8 @@ class TelegramBridgeConfig:
     voice_transcription: bool = False
     voice_max_bytes: int = 10 * 1024 * 1024
     voice_transcription_model: str = "gpt-4o-mini-transcribe"
+    voice_transcription_base_url: str | None = None
+    voice_transcription_api_key: str | None = None
     forward_coalesce_s: float = 1.0
     media_group_debounce_s: float = 1.0
     files: TelegramFilesSettings = field(default_factory=TelegramFilesSettings)

src/takopi/telegram/loop.py

@@ -1663,6 +1663,8 @@ async def run_main_loop(
                         model=cfg.voice_transcription_model,
                         max_bytes=cfg.voice_max_bytes,
                         reply=reply,
+                        base_url=cfg.voice_transcription_base_url,
+                        api_key=cfg.voice_transcription_api_key,
                     )
                     if text is None:
                         return

src/takopi/telegram/voice.py

@@ -28,10 +28,23 @@ class VoiceTranscriber(Protocol):
 
 
 class OpenAIVoiceTranscriber:
+    def __init__(
+        self,
+        *,
+        base_url: str | None = None,
+        api_key: str | None = None,
+    ) -> None:
+        self._base_url = base_url
+        self._api_key = api_key
+
     async def transcribe(self, *, model: str, audio_bytes: bytes) -> str:
         audio_file = io.BytesIO(audio_bytes)
         audio_file.name = "voice.ogg"
-        async with AsyncOpenAI(timeout=120) as client:
+        async with AsyncOpenAI(
+            base_url=self._base_url,
+            api_key=self._api_key,
+            timeout=120,
+        ) as client:
             response = await client.audio.transcriptions.create(
                 model=model,
                 file=audio_file,
@@ -48,6 +61,8 @@ async def transcribe_voice(
     max_bytes: int | None = None,
     reply: Callable[..., Awaitable[None]],
     transcriber: VoiceTranscriber | None = None,
+    base_url: str | None = None,
+    api_key: str | None = None,
 ) -> str | None:
     voice = msg.voice
     if voice is None:
@@ -74,7 +89,7 @@ async def transcribe_voice(
         await reply(text="voice message is too large to transcribe.")
         return None
     if transcriber is None:
-        transcriber = OpenAIVoiceTranscriber()
+        transcriber = OpenAIVoiceTranscriber(base_url=base_url, api_key=api_key)
     try:
         return await transcriber.transcribe(model=model, audio_bytes=audio_bytes)
     except OpenAIError as exc:

tests/test_cli_helpers.py

@@ -135,6 +135,23 @@ def test_doctor_voice_checks(monkeypatch) -> None:
     monkeypatch.delenv("OPENAI_API_KEY", raising=False)
     checks = cli._doctor_voice_checks(settings)
     assert checks[0].status == "error"
+    assert checks[0].detail == "API key not set"
+
+    settings_with_key = _settings(
+        {
+            "transports": {
+                "telegram": {
+                    "bot_token": "token",
+                    "chat_id": 1,
+                    "voice_transcription": True,
+                    "voice_transcription_api_key": "local",
+                }
+            }
+        }
+    )
+    checks = cli._doctor_voice_checks(settings_with_key)
+    assert checks[0].status == "ok"
+    assert checks[0].detail == "voice_transcription_api_key set"
 
     monkeypatch.setenv("OPENAI_API_KEY", "key")
     checks = cli._doctor_voice_checks(settings)

tests/test_telegram_backend.py

@@ -143,6 +143,8 @@ def test_telegram_backend_build_and_run_wires_config(
         voice_transcription=True,
         voice_max_bytes=1234,
         voice_transcription_model="whisper-1",
+        voice_transcription_base_url="http://localhost:8000/v1",
+        voice_transcription_api_key="local",
         files=TelegramFilesSettings(enabled=True, allowed_user_ids=[1, 2]),
         topics=TelegramTopicsSettings(enabled=True, scope="main"),
     )
@@ -161,6 +163,8 @@ def test_telegram_backend_build_and_run_wires_config(
     assert cfg.voice_transcription is True
     assert cfg.voice_max_bytes == 1234
     assert cfg.voice_transcription_model == "whisper-1"
+    assert cfg.voice_transcription_base_url == "http://localhost:8000/v1"
+    assert cfg.voice_transcription_api_key == "local"
     assert cfg.files.enabled is True
     assert cfg.files.allowed_user_ids == [1, 2]
     assert cfg.topics.enabled is True

tests/test_telegram_bridge.py

@@ -1954,8 +1954,10 @@ async def test_run_main_loop_voice_transcript_preserves_directive(
         model: str,
         max_bytes: int | None = None,
         reply,
+        base_url: str | None = None,
+        api_key: str | None = None,
     ) -> str:
-        _ = bot, msg, enabled, model, max_bytes, reply
+        _ = bot, msg, enabled, model, max_bytes, reply, base_url, api_key
         return "/codex do thing"
 
     monkeypatch.setattr(telegram_loop, "transcribe_voice", _fake_transcribe)