Add startup model preparation endpoint and update Whisper handling

backend/main.py

@@ -9,7 +9,7 @@ import json
 from . import models, schemas
 from .database import Base, engine, SessionLocal, ensure_sources_column
 from .local_rag import router as local_rag_router
-from .ollama_admin import inspect_ollama_startup, pull_local_model, start_local_ollama
+from .ollama_admin import inspect_ollama_startup, prepare_startup_models, pull_local_model, start_local_ollama
 from .ollama_client import list_models as ollama_list, chat as ollama_chat, chat_stream as ollama_chat_stream
 from .websearch import enrich_prompt
 
@@ -73,6 +73,16 @@ async def ollama_pull_route(req: schemas.OllamaPullRequest):
     except RuntimeError as exc:
         raise HTTPException(status_code=400, detail=str(exc)) from exc
 
+
+@app.post("/startup/prepare-models")
+async def startup_prepare_models_route():
+    try:
+        return await prepare_startup_models()
+    except FileNotFoundError as exc:
+        raise HTTPException(status_code=404, detail=str(exc)) from exc
+    except RuntimeError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
 @app.get("/sessions", response_model=schemas.SessionsResponse)
 def get_sessions(db: Session = Depends(get_db)):
     sessions = db.query(models.ChatSession).order_by(models.ChatSession.created_at.desc()).all()

backend/rag/corpus_builder.py

@@ -69,6 +69,8 @@ import faulthandler, signal
 import multiprocessing as mp_context
 import time
 
+from backend.whisper_admin import ensure_whisper_model_downloaded, whisper_runtime_error as whisper_admin_runtime_error
+
 os.environ.setdefault("PYTORCH_ENABLE_MPS_FALLBACK", "1")
 os.environ.setdefault("OBJC_DISABLE_INITIALIZE_FORK_SAFETY", "YES")
 
@@ -1431,12 +1433,7 @@ def _load_torch_module():
         return None
 
 def whisper_runtime_error() -> Optional[str]:
-    if importlib.util.find_spec("whisper") is None:
-        return (
-            "Audio/video transcription requires the optional 'openai-whisper' package. "
-            "Install it in backend/.venv, for example: pip install -U openai-whisper"
-        )
-    return None
+    return whisper_admin_runtime_error()
 
 def _resolve_whisper_device(flag: str) -> Optional[str]:
     if flag and flag != "auto":
@@ -1495,6 +1492,8 @@ def process_media(path: Path, args) -> List[Record]:
     if whisper_error:
         raise RuntimeError(whisper_error)
 
+    ensure_whisper_model_downloaded(args.whisper_model)
+
     ffmpeg_bin = resolve_binary_path(args.ffmpeg, "HEIMGEIST_FFMPEG_PATH", "ffmpeg", "/usr/bin/ffmpeg")
     ffprobe_bin = resolve_binary_path(args.ffprobe, "HEIMGEIST_FFPROBE_PATH", "ffprobe", "/usr/bin/ffprobe")
     missing_tools = []