transcript caching

Dockerfile

@@ -18,6 +18,12 @@ RUN apt-get update && \
     rm -rf /tmp/deno.zip && \
     rm -rf /var/lib/apt/lists/*
 
+# populate cache; this works around HuggingFace egress restrictions
+RUN curl -fsSL "https://ndurner.de/download/aileen3/aileen3-cache.zip" -o /tmp/aileen3-cache.zip && \
+    unzip -q /tmp/aileen3-cache.zip -d /tmp && \
+    mv /tmp/aileen3-cache /root/.cache/aileen3 && \
+    rm -rf /tmp/aileen3-cache.zip
+
 COPY demo/requirements.txt demo/requirements.txt
 COPY mcp mcp
 RUN pip install --upgrade pip && \

demo/problem_cell.py

@@ -1,5 +1,8 @@
 from __future__ import annotations
 
+import hashlib
+import json
+import os
 import tempfile
 from pathlib import Path
 from urllib.parse import parse_qs, urlparse
@@ -22,6 +25,13 @@ HERE = Path(__file__).parent
 ASSETS_DIR = HERE / "assets"
 DIGITALGIPFEL_IMG = ASSETS_DIR / "digitalgipfel.jpeg"
 
+BASE_CACHE = Path(os.environ.get("AILEEN3_CACHE_DIR", Path.home() / ".cache" / "aileen3"))
+TRANSCRIPTION_CACHE = BASE_CACHE / "transcription"
+
+
+def _transcription_cache_path(reference: str) -> Path:
+    return TRANSCRIPTION_CACHE / f"{reference}.json"
+
 
 def render_status_box(message: str, tone: str = "placeholder") -> str:
     tone_class = {
@@ -44,11 +54,26 @@ def _extract_video_id(video_url: str) -> str | None:
 
 
 def _fetch_transcript(video_url: str) -> tuple[str | None, str | None]:
+    TRANSCRIPTION_CACHE.mkdir(parents=True, exist_ok=True)
+
     if YoutubeDL is None:  # pragma: no cover - dependency should always be present
         return None, "yt-dlp is not installed in this environment."
     video_id = _extract_video_id(video_url)
     if not video_id:
         return None, "That does not look like a valid YouTube URL with a video id."
+
+    # Align cache layout with `media_tools`: transcription cache under BASE_CACHE/transcription
+    # using a stable reference derived from the YouTube video id when available.
+    reference = f"youtube_{hashlib.sha256(video_id.encode('utf-8')).hexdigest()[:32]}"
+    cache_path = _transcription_cache_path(reference)
+    if cache_path.exists():
+        try:
+            cached = json.loads(cache_path.read_text(encoding="utf-8"))
+        except Exception:
+            cached = None
+        if isinstance(cached, str) and cached.strip():
+            return cached, None
+
     with tempfile.TemporaryDirectory() as tmpdir:
         output_template = str(Path(tmpdir) / "%(id)s.%(ext)s")
         ydl_opts = {
@@ -84,6 +109,13 @@ def _fetch_transcript(video_url: str) -> tuple[str | None, str | None]:
     readable = " ".join(text_chunks).strip()
     if not readable:
         return None, "Transcript was empty. Try again or choose another video."
+
+    try:
+        cache_path.write_text(json.dumps(readable), encoding="utf-8")
+    except Exception:
+        # Cache failures should not block the happy path.
+        pass
+
     return readable, None