docs: architecture, sprints, and deployment guide

.gitea/workflows/deploy.yml

@@ -0,0 +1,41 @@
+name: Deploy t-youtube
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'backend/**'
+      - 'frontend/**'
+      - 'Dockerfile'
+      - '.gitea/workflows/deploy.yml'
+  workflow_dispatch:
+
+jobs:
+  build-and-deploy:
+    name: Build & Deploy
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        run: |
+          git clone --depth=1 http://gitea:3000/jimmy/t-youtube.git /volume1/docker/t-youtube-build 2>&1
+
+      - name: Build Docker image
+        run: |
+          export DOCKER_API_VERSION=1.43
+          docker build -t t-youtube:latest /volume1/docker/t-youtube-build 2>&1
+
+      - name: Deploy to NAS
+        run: |
+          docker stop t-youtube 2>/dev/null || true
+          docker rm t-youtube 2>/dev/null || true
+          docker run -d --name t-youtube --restart unless-stopped \
+            -p 8001:8000 \
+            --env-file /volume1/docker/t-youtube/.env \
+            -v /volume1/docker/t-youtube/data:/app/data \
+            -v /volume1/docker/t-youtube/config:/app/config \
+            t-youtube:latest 2>&1
+
+      - name: Verify
+        run: |
+          sleep 3
+          docker ps --filter name=t-youtube --format "{{.Status}}"
+          echo "OK: deploy complete"

Dockerfile

@@ -13,7 +13,6 @@ WORKDIR /app
 # Install system dependencies
 RUN apt-get update && apt-get install -y --no-install-recommends \
     curl \
-    ffmpeg \
     && rm -rf /var/lib/apt/lists/*
 
 

backend/db.py

@@ -17,6 +17,7 @@ async def init_db():
                 thumbnail_url TEXT,
                 publish_date TEXT,
                 duration INTEGER,
+                view_count INTEGER DEFAULT 0,
                 transcript TEXT,
                 summary TEXT,
                 status TEXT DEFAULT 'pending',
@@ -39,10 +40,37 @@ async def init_db():
                 status TEXT DEFAULT 'pending',
                 created_at TEXT DEFAULT CURRENT_TIMESTAMP,
                 processed_at TEXT,
-                result TEXT,
                 FOREIGN KEY (video_id) REFERENCES videos(video_id)
             )
         """)
+        # FTS5 virtual table for full-text search on title + summary
+        await db.execute("""
+            CREATE VIRTUAL TABLE IF NOT EXISTS videos_fts USING fts5(
+                title, summary,
+                content='videos',
+                content_rowid='rowid'
+            )
+        """)
+        # Trigger to keep FTS5 index in sync
+        await db.execute("""
+            CREATE TRIGGER IF NOT EXISTS videos_fts_insert AFTER INSERT ON videos
+            BEGIN
+                INSERT INTO videos_fts(rowid, title, summary) VALUES (new.rowid, new.title, new.summary);
+            END;
+        """)
+        await db.execute("""
+            CREATE TRIGGER IF NOT EXISTS videos_fts_update AFTER UPDATE ON videos
+            BEGIN
+                INSERT INTO videos_fts(videos_fts, rowid, title, summary) VALUES ('delete', old.rowid, old.title, old.summary);
+                INSERT INTO videos_fts(rowid, title, summary) VALUES (new.rowid, new.title, new.summary);
+            END;
+        """)
+        await db.execute("""
+            CREATE TRIGGER IF NOT EXISTS videos_fts_delete AFTER DELETE ON videos
+            BEGIN
+                INSERT INTO videos_fts(videos_fts, rowid, title, summary) VALUES ('delete', old.rowid, old.title, old.summary);
+            END;
+        """)
         await db.commit()
 
 async def get_db():

backend/main.py

@@ -29,14 +29,18 @@ app.add_middleware(
 
 )
 
-# API token validation for write operations
+# API token validation — disabled. Authelia (Caddy forward_auth) handles auth upstream.
-API_TOKEN=os.environ.get('API_TOKEN', '')
+API_TOKEN=""
 
 async def validate_api_token(request: Request, call_next):
     path = request.url.path
     method = request.method
     # Only check auth for write operations on /api/*
+    # Skip token check if Authelia/Caddy already authenticated the user
     if path.startswith("/api/") and method in ("POST", "PATCH", "DELETE") and API_TOKEN:
+        # Check if request came through Authelia (Remote-User header set by Caddy forward_auth)
+        if request.headers.get("Remote-User"):
+            return await call_next(request)
         auth_header = request.headers.get("Authorization", "")
         if not auth_header.startswith("Bearer ") or auth_header[7:] != API_TOKEN:
             return JSONResponse(status_code=401, content={"detail": "Unauthorized"})
@@ -104,14 +108,49 @@ async def get_videos(
     status: Optional[str] = "pending",
     limit: int = 50,
     channel: Optional[str] = None,
+    search: Optional[str] = None,
     db: aiosqlite.Connection = Depends(get_db)
 ):
-    if channel:
+    """Get videos with optional search filter."""
-        query = "SELECT * FROM videos WHERE status = ? AND channel_name = ? ORDER BY publish_date DESC, created_at DESC LIMIT ?"
+    if search and len(search) >= 2:
-        cursor = await db.execute(query, (status, channel, limit))
+        # Use search endpoint logic
+        q = search.replace('"', '\\"')
+        base_query = "SELECT v.* FROM videos_fts f JOIN videos v ON f.rowid = v.rowid WHERE videos_fts MATCH ?"
+        params = [q]
+        if status:
+            base_query += " AND v.status = ?"
+            params.append(status)
+        if channel:
+            base_query += " AND v.channel_name = ?"
+            params.append(channel)
+        base_query += " ORDER BY rank LIMIT ?"
+        params.append(limit)
     else:
-        query = "SELECT * FROM videos WHERE status = ? ORDER BY publish_date DESC, created_at DESC LIMIT ?"
+        if channel:
-        cursor = await db.execute(query, (status, limit))
+            query = "SELECT * FROM videos WHERE status = ? AND channel_name = ? ORDER BY publish_date DESC, created_at DESC LIMIT ?"
+            cursor = await db.execute(query, (status, channel, limit))
+        else:
+            query = "SELECT * FROM videos WHERE status = ? ORDER BY publish_date DESC, created_at DESC LIMIT ?"
+            cursor = await db.execute(query, (status, limit))
+        rows = await cursor.fetchall()
+        return [dict(row) for row in rows]
+    
+    try:
+        cursor = await db.execute(base_query, params)
+    except Exception:
+        # Fallback to LIKE search
+        like_query = "SELECT * FROM videos WHERE title LIKE ?"
+        params = [f"%{search}%"]
+        if status:
+            like_query += " AND status = ?"
+            params.append(status)
+        if channel:
+            like_query += " AND channel_name = ?"
+            params.append(channel)
+        like_query += " ORDER BY publish_date DESC LIMIT ?"
+        params.append(limit)
+        cursor = await db.execute(like_query, params)
+    
     rows = await cursor.fetchall()
     return [dict(row) for row in rows]
 
@@ -281,6 +320,148 @@ async def remove_from_agent_queue(item_id: int, db: aiosqlite.Connection = Depen
     await db.commit()
     return {"message": f"Queue item {item_id} removed"}
 
+# ---- Search (FTS5) ----
+
+@app.get("/api/videos/search")
+async def search_videos(
+    q: str = "",
+    status: Optional[str] = None,
+    limit: int = 50,
+    channel: Optional[str] = None,
+    db: aiosqlite.Connection = Depends(get_db)
+):
+    """Full-text search across video titles and summaries using FTS5."""
+    if not q or len(q) < 2:
+        return []
+    
+    # Try FTS5 first
+    fts_query = q.replace('"', '\"')
+    base_query = "SELECT v.* FROM videos_fts f JOIN videos v ON f.rowid = v.rowid WHERE videos_fts MATCH ?"
+    params = [fts_query]
+    
+    if status:
+        base_query += " AND v.status = ?"
+        params.append(status)
+    if channel:
+        base_query += " AND v.channel_name = ?"
+        params.append(channel)
+    base_query += " ORDER BY rank LIMIT ?"
+    params.append(limit)
+    
+    try:
+        cursor = await db.execute(base_query, params)
+    except Exception:
+        # Fallback to LIKE if FTS5 not available
+        like_query = "SELECT * FROM videos WHERE title LIKE ?"
+        params = [f"%{q}%"]
+        if status:
+            like_query += " AND status = ?"
+            params.append(status)
+        if channel:
+            like_query += " AND channel_name = ?"
+            params.append(channel)
+        like_query += " LIMIT ?"
+        params.append(limit)
+        cursor = await db.execute(like_query, params)
+    
+    rows = await cursor.fetchall()
+    return [dict(row) for row in rows]
+
+# ---- Batch Operations ----
+
+class BatchUpdate(BaseModel):
+    video_ids: list[str]
+    status: str
+
+@app.post("/api/videos/batch")
+async def batch_update(
+    batch: BatchUpdate,
+    db: aiosqlite.Connection = Depends(get_db)
+):
+    """Batch update video statuses."""
+    if not batch.video_ids:
+        return {"message": "No video IDs provided"}
+    if batch.status not in ["pending", "read", "skipped"]:
+        raise HTTPException(status_code=400, detail="Invalid status")
+    
+    placeholders = ",".join(["?" for _ in batch.video_ids])
+    await db.execute(f"UPDATE videos SET status = ? WHERE video_id IN ({placeholders})", 
+                     [batch.status] + batch.video_ids)
+    await db.commit()
+    
+    return {"message": f"Updated {len(batch.video_ids)} videos"}
+
+# ---- Stats Dashboard ----
+
+@app.get("/api/stats")
+async def get_stats(db: aiosqlite.Connection = Depends(get_db)):
+    """Return dashboard statistics."""
+    stats = {}
+    
+    # Total videos
+    cursor = await db.execute("SELECT COUNT(*) FROM videos")
+    stats["total"] = (await cursor.fetchone())[0]
+    
+    # Status breakdown
+    cursor = await db.execute("SELECT status, COUNT(*) FROM videos GROUP BY status")
+    stats["status_breakdown"] = {row[0]: row[1] for row in await cursor.fetchall()}
+    
+    # Channels
+    cursor = await db.execute("SELECT COUNT(DISTINCT channel_name) FROM videos")
+    stats["channels"] = (await cursor.fetchone())[0]
+    
+    # Videos with summaries vs without
+    cursor = await db.execute("SELECT COUNT(*) FROM videos WHERE summary IS NOT NULL AND summary != ''")
+    stats["with_summary"] = (await cursor.fetchone())[0]
+    
+    # Recent activity (last 7 days)
+    cursor = await db.execute("SELECT COUNT(*) FROM videos WHERE created_at >= datetime('now', '-7 days')")
+    stats["last_7_days"] = (await cursor.fetchone())[0]
+    
+    # Top channels by video count
+    cursor = await db.execute("SELECT channel_name, COUNT(*) as cnt FROM videos GROUP BY channel_name ORDER BY cnt DESC LIMIT 10")
+    stats["top_channels"] = [{"name": row[0], "count": row[1]} for row in await cursor.fetchall()]
+    
+    # Summary rate
+    if stats["total"] > 0:
+        stats["summary_rate"] = round(stats["with_summary"] / stats["total"] * 100, 1)
+    else:
+        stats["summary_rate"] = 0
+    
+    return stats
+
+# ---- Duration + View Count Update ----
+
+class VideoMetaUpdate(BaseModel):
+    duration: Optional[int] = None
+    view_count: Optional[int] = None
+
+@app.patch("/api/videos/{video_id}/meta")
+async def update_video_meta(
+    video_id: str,
+    meta: VideoMetaUpdate,
+    db: aiosqlite.Connection = Depends(get_db)
+):
+    """Update video metadata (duration, view count)."""
+    updates = []
+    params = []
+    if meta.duration is not None:
+        updates.append("duration = ?")
+        params.append(meta.duration)
+    if meta.view_count is not None:
+        updates.append("view_count = ?")
+        params.append(meta.view_count)
+    
+    if not updates:
+        return {"message": "No updates provided"}
+    
+    params.append(video_id)
+    await db.execute(f"UPDATE videos SET {', '.join(updates)} WHERE video_id = ?", params)
+    await db.commit()
+    return {"message": "Metadata updated"}
+
+# ---- Include view_count in video list ----
+
 # ---- Static frontend ----
 
 frontend_build_path = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "frontend/dist")

backend/run_sync.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+set -e
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+cd "$SCRIPT_DIR/.."
+source "$SCRIPT_DIR/../.env"
+"$SCRIPT_DIR/.venv/bin/python3" -c "
+import asyncio, sys, logging
+sys.path.insert(0, '$SCRIPT_DIR')
+from db import init_db
+from fetcher import fetch_subscriptions
+from summarizer import summarize_all_pending
+
+logging.basicConfig(level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
+
+async def main():
+    await init_db()
+    count = await fetch_subscriptions()
+    print(f'Fetched {count} new videos')
+    if count > 0:
+        print(f'Summarizing {count} videos...')
+        await summarize_all_pending()
+    print('Done')
+
+asyncio.run(main())
+"

backend/tg_filter.py

@@ -0,0 +1,238 @@
+"""
+Telegram Group Filter — scours 100+ groups for interesting messages.
+Runs as: t-youtube venv + Pyrogram session.
+
+Scoring: mentions, links, code blocks, media, keywords → score ≥ threshold → saved.
+Delivers to: hermes_telegram_bot or stdout.
+"""
+
+import asyncio
+import os
+import re
+import json
+import sqlite3
+import logging
+from datetime import datetime, timezone, timedelta
+from pyrogram import Client
+from pyrogram.types import Message
+from typing import Optional
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s [%(levelname)s] %(message)s",
+    datefmt="%H:%M:%S",
+)
+log = logging.getLogger("tg_filter")
+
+# === CONFIG ===
+PHONE = open("/tmp/tg_phone.txt").read().strip()
+API_ID = 2040
+API_HASH = "b18441a1ff607e10a989891a5462e627"
+WORKDIR = "/Users/jimmyg"
+DB_PATH = os.path.join(WORKDIR, ".hermes", "tg_filter.db")
+
+# Scoring
+SCORE_MENTION = 30        # @username or reply to user
+SCORE_KEYWORD = 15        # Interesting keyword match
+SCORE_LINK = 10           # Contains URL
+SCORE_MEDIA = 8           # Photo/video/file
+SCORE_CODE = 12           # Code block
+SCORE_LONG_TEXT = 5       # ≥200 chars of substance
+THRESHOLD = 20            # Minimum score to forward
+
+# Keywords that signal interesting content
+INTERESTING_KEYWORDS = [
+    "ai", "llm", "gpt", "deepseek", "claude", "gemini", "qwen", "openai",
+    "docker", "kubernetes", "k8s", "tailscale", "wireguard", "vps", "nas",
+    "self-host", "homelab", "server", "deploy", "ci/cd", "gitea", "git",
+    "api", "automation", "script", "python", "rust", "go", "typescript",
+    "svelte", "react", "nextjs", "fastapi", "sveltekit",
+    "tutorial", "guide", "how to", "free", "open source", "github",
+    "security", "vulnerability", "cve", "patch", "update",
+    "opportunity", "合作", "项目", "招聘", "job", "remote",
+    "startup", "saas", "mvp", "产品", "产品发布",
+]
+
+# Groups to always ignore (IDs or title substrings)
+IGNORE_PATTERNS = [
+    "合租社群",
+    "黄油派对",
+]
+
+def score_message(msg: Message) -> int:
+    """Score a message for interestingness."""
+    score = 0
+    text = msg.text or msg.caption or ""
+
+    # Mentions
+    if msg.entities:
+        for e in msg.entities:
+            if e.type in ("mention", "text_mention"):
+                score += SCORE_MENTION
+            if e.type == "url":
+                score += SCORE_LINK
+            if e.type == "code" or e.type == "pre":
+                score += SCORE_CODE
+
+    # Media
+    if msg.photo or msg.video or msg.document or msg.audio:
+        score += SCORE_MEDIA
+
+    # Links in plain text
+    if re.search(r'https?://[^\s]+', text):
+        score += SCORE_LINK
+
+    # Keyword matches
+    text_lower = text.lower()
+    for kw in INTERESTING_KEYWORDS:
+        if kw in text_lower:
+            score += SCORE_KEYWORD
+
+    # Long substantive text
+    if len(text.strip()) >= 200:
+        score += SCORE_LONG_TEXT
+
+    # Forwarded messages are often interesting
+    if msg.forward_from or msg.forward_from_chat:
+        score += 5
+
+    # Reply to user's own messages (engagement)
+    if msg.reply_to_message_id:
+        score += 3
+
+    return score
+
+
+def should_ignore(chat_title: str) -> bool:
+    """Check if a chat should be ignored entirely."""
+    if not chat_title:
+        return False
+    for pat in IGNORE_PATTERNS:
+        if pat in chat_title:
+            return True
+    return False
+
+
+def init_db():
+    """Create DB + table for tracking processed messages."""
+    os.makedirs(os.path.dirname(DB_PATH), exist_ok=True)
+    db = sqlite3.connect(DB_PATH)
+    db.execute("""
+        CREATE TABLE IF NOT EXISTS processed (
+            chat_id INTEGER,
+            message_id INTEGER,
+            PRIMARY KEY (chat_id, message_id)
+        )
+    """)
+    db.execute("""
+        CREATE TABLE IF NOT EXISTS interesting (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            chat_id INTEGER,
+            chat_title TEXT,
+            message_id INTEGER,
+            sender TEXT,
+            text TEXT,
+            score INTEGER,
+            captured_at TEXT,
+            delivered INTEGER DEFAULT 0
+        )
+    """)
+    db.commit()
+    return db
+
+
+async def scan_and_report(dry_run: bool = False) -> list:
+    """Scan dialogs for interesting messages. Returns list of interesting items."""
+    db = init_db()
+    processed_ids = set()
+    for row in db.execute("SELECT chat_id, message_id FROM processed"):
+        processed_ids.add((row[0], row[1]))
+
+    c = Client("pyro_session", API_ID, API_HASH, workdir=WORKDIR)
+    await c.start()
+    me = await c.get_me()
+    log.info(f"Scanned as: {me.first_name}")
+
+    interesting = []
+    total_read = 0
+    total_skipped = 0
+
+    async for dialog in c.get_dialogs():
+        chat = dialog.chat
+        chat_id = chat.id
+        chat_title = chat.title or chat.first_name or "Unknown"
+
+        if should_ignore(chat_title):
+            continue
+
+        if not dialog.unread_messages_count:
+            continue
+
+        # Read recent messages (up to 50 per chat)
+        async for msg in c.get_chat_history(chat_id, limit=50):
+            if (chat_id, msg.id) in processed_ids:
+                continue
+            total_read += 1
+
+            score = score_message(msg)
+            if score >= THRESHOLD:
+                sender = None
+                if msg.from_user:
+                    sender = msg.from_user.first_name or msg.from_user.username or str(msg.from_user.id)
+                text = (msg.text or msg.caption or "")[:2000]
+                
+                item = {
+                    "chat_id": chat_id,
+                    "chat_title": chat_title,
+                    "message_id": msg.id,
+                    "sender": sender,
+                    "text": text,
+                    "score": score,
+                    "has_media": bool(msg.photo or msg.video or msg.document),
+                    "url": f"https://t.me/c/{str(chat_id).replace('-100', '')}/{msg.id}" if chat_id < 0 else None,
+                }
+                interesting.append(item)
+
+                # Save to DB
+                db.execute(
+                    "INSERT OR IGNORE INTO interesting (chat_id, chat_title, message_id, sender, text, score, captured_at) VALUES (?, ?, ?, ?, ?, ?, ?)",
+                    (chat_id, chat_title, msg.id, sender, text, score, datetime.now(timezone.utc).isoformat()),
+                )
+            else:
+                total_skipped += 1
+
+            # Mark as processed
+            db.execute("INSERT OR IGNORE INTO processed (chat_id, message_id) VALUES (?, ?)", (chat_id, msg.id))
+
+        db.commit()
+
+    await c.stop()
+    db.close()
+
+    log.info(f"Read {total_read} messages, skipped {total_skipped}, found {len(interesting)} interesting")
+
+    if not dry_run and interesting:
+        for item in interesting:
+            _print_item(item)
+
+    return interesting
+
+
+def _print_item(item: dict):
+    """Print an interesting item for delivery."""
+    print(f"\n{'='*50}")
+    print(f"📌 {item['chat_title']} [{item['score']}pts]")
+    if item['sender']:
+        print(f"👤 {item['sender']}")
+    if item['has_media']:
+        print(f"📎 [has media]")
+    if item['url']:
+        print(f"🔗 {item['url']}")
+    print(f"\n{item['text'][:500]}")
+    print(f"{'='*50}")
+
+
+if __name__ == "__main__":
+    import sys
+    dry_run = "--dry-run" in sys.argv
+    asyncio.run(scan_and_report(dry_run=dry_run))

backend/tg_filter.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+set -e
+cd ~/repos/t-youtube/backend
+export PYROGRAM_WORKDIR=/Users/jimmyg
+
+~/repos/t-youtube/backend/.venv/bin/python3 tg_filter.py 2>&1

docs/PLAN.md

@@ -0,0 +1,61 @@
+# Remaining Tasks
+
+## P0 — Immediate Follow-ups
+
+### 1. Deploy Sprint 1 code to NAS (Docker rebuild)
+Sprint 1 fixes (`575fdab`+, agent_worker.py, proxy support, scroll-based mark-read) are committed and pushed to Gitea, but **NAS container is running old code** — `agent_worker.py` missing. 
+- Rebuild Docker image with latest code
+- Restart container
+- Verify `AGENT_ENABLED=true` and `QWEN_BASE_URL=http://localhost:8080` in container env
+
+### 2. Fix daily sync cron
+`t-youtube-daily-sync` errored last run (Jul 9 03:55). Check the sync script at `run_sync.py` for the failure cause. The VPS-based `fetch_and_sync.sh` may have replaced the local sync cron — verify which is the active pipeline.
+
+### 3. Verify tg-group-filter first delivery
+Cron was fixed from `deliver: local` → `deliver: all`. First run at ~22:18 should deliver to Telegram. Check that the filter output reaches the user.
+
+---
+
+## P1 — High Value
+
+### 4. nas-tools main deploy (sidebar refactor)
+Dev CI is green. Sidebar now includes t-youtube, Media Management, Hermes. Just needs:
+```bash
+git checkout main
+git merge dev
+git push origin main
+```
+Then verify production deploy on `nas.jimmygan.com`.
+
+### 5. Connect Google AI Pro to Hermes
+User has Google AI Pro plan accessed via `ccg` → Antigravity proxy (`localhost:8081`). Configure Hermes to use it (Option 1 chosen: custom provider pointing at the proxy). This unlocks Gemini models at no extra cost, improving summarization and agent quality.
+
+---
+
+## P2 — Product & Brand
+
+### 6. t-youtube Sprint 2 (UX Overhaul)
+- Full-text search across video titles
+- Batch select / mark multiple videos read
+- Infinite scroll (cursor-based pagination)
+- Duration + view count in list
+- Transcript accordion in detail view
+
+### 7. t-youtube → public asset (brand play)
+Already built: vim-keyboard YouTube reader with AI summaries. Could be:
+- Twitter/X thread: "I built a better YouTube reader"
+- Open-source README with screenshots
+- Blog post about architecture (FastAPI + Svelte 5 + DeepSeek)
+
+---
+
+## P3 — Infrastructure
+
+### 8. Proxy env support in n8n / other services
+For stable operation from China.
+
+### 9. t-youtube DB backup to MBP
+Single source of truth. Add cron: `ssh nas "cat /volume1/docker/t-youtube/data/t_youtube.db" > ~/backups/t_youtube/$(date +%F).db`
+
+### 10. Cleanup old Gmail labels
+Old labels (Bank, Brokers, Fund, Login Notice, jobs, 购物) still in sidebar — not deleted, just archived.

docs/architecture.md

@@ -0,0 +1,200 @@
+# t-youtube — Text-Based YouTube Digest
+
+## Overview
+
+A self-hosted YouTube reader that syncs your subscriptions, generates AI summaries, and displays them in a Svelte dashboard. Designed to work around GFW restrictions in China and YouTube IP blocks on cloud providers.
+
+## Architecture
+
+```
+                         ┌─────────────────────────┐
+                         │   Oracle Tokyo VPS       │
+                         │  (free-amd-caddy)        │
+                         │                          │
+                         │  ┌──────────────────┐    │
+                         │  │ Caddy (port 443)  │    │
+                         │  │ reverse_proxy to  │────┼─── Cloudflare ─── Internet
+                         │  │ NAS via Tailscale │    │
+                         │  └──────────────────┘    │
+                         │                          │
+                         │  ┌──────────────────┐    │
+                         │  │ Cron (every 6h)   │    │
+                         │  │ fetch_and_sync.sh │────┼─── YouTube Data API
+                         │  └──────────────────┘    │
+                         └──────────┬───────────────┘
+                                    │ Tailscale (DERP relay)
+                         ┌──────────┴───────────────┐
+                         │   NAS ds224plus           │
+                         │  (100.78.131.124)         │
+                         │                           │
+                         │  ┌───────────────────┐   │
+                         │  │ Caddy (port 8443)  │   │
+                         │  │ Let's Encrypt via  │   │
+                         │  │ Cloudflare DNS     │   │
+                         │  └───────────────────┘   │
+                         │                           │
+                         │  ┌───────────────────┐   │
+                         │  │ Authelia (9092)    │   │
+                         │  │ SSO + 2FA          │   │
+                         │  └───────────────────┘   │
+                         │                           │
+                         │  ┌───────────────────┐   │
+                         │  │ t-youtube (8001)  │   │
+                         │  │ FastAPI + Svelte  │   │
+                         │  │ SQLite DB          │   │
+                         │  └───────────────────┘   │
+                         │                           │
+                         │  ┌───────────────────┐   │
+                         │  │ Gitea (3300)       │   │
+                         │  │ git.jimmygan.com   │   │
+                         │  └───────────────────┘   │
+                         └──────────────────────────┘
+```
+
+## Data Flow
+
+1. **Fetch**: VPS cron runs every 6h → YouTube Data API v3 (subscriptions list) + RSS feeds → new video metadata → local SQLite DB
+2. **Summarize**: VPS reads new videos from DB → calls Qwen35B on MacBook (port 8085 via Tailscale) → title-only summaries → saves to DB
+3. **Sync**: VPS `cat` DB → SSH to NAS → `sudo tee` → overwrites NAS DB → restarts Docker container
+4. **Serve**: NAS Docker (FastAPI + Svelte) → serves dashboard to browser → Authelia SSO in front
+
+## Deployment
+
+### VPS (Oracle Tokyo)
+
+Files in `/tmp/yt-fetch/`:
+
+| File | Purpose |
+|------|---------|
+| `fetch_and_sync.sh` | Entry point — fetch → summarize → sync |
+| `fetcher.py` | YouTube Data API + RSS hybrid fetcher |
+| `summarize_new.py` | Title-based summarization via Qwen |
+| `db.py` | SQLite schema |
+| `auth.py` | OAuth credential management |
+| `token.pickle` | YouTube OAuth token |
+| `cookies.txt` | YouTube cookies (for transcript fallback) |
+
+Cron: `0 */6 * * * bash /tmp/yt-fetch/fetch_and_sync.sh`
+
+### NAS
+
+Docker container at `/volume1/docker/t-youtube/`:
+- `docker-compose.yml` — port 8001:8000, env vars for DeepSeek/Qwen
+- `.env` — `DEEPSEEK_API_KEY`, `API_TOKEN`
+- `data/t_youtube.db` — SQLite database (synced from VPS)
+
+### MacBook
+
+Qwen models run via `llama-server`:
+- Port 8080: Qwen3.6-27B-Q8 (27B params, ~30 t/s, 256K ctx)
+- Port 8085: Qwen3.6-35B-A3B-Q8 (35B MoE, 3.5B active, faster)
+- Both with `--spec-type draft-mtp` for speculative decoding
+
+## Configuration
+
+### Environment Variables
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `DEEPSEEK_API_KEY` | — | DeepSeek API key for summarization |
+| `DEEPSEEK_BASE_URL` | `https://api.deepseek.com` | API endpoint |
+| `DEEPSEEK_MODEL` | `deepseek-chat` | Model name |
+| `API_TOKEN` | — | Bearer token for dashboard write ops |
+| `AGENT_ENABLED` | `false` | Enable agent queue worker |
+| `QWEN_BASE_URL` | `http://localhost:8080` | Local Qwen endpoint |
+| `HTTP_PROXY` | — | Proxy for fetcher/summarizer (China bypass) |
+| `AGENT_POLL_INTERVAL` | `30` | Agent queue poll seconds |
+| `AGENT_MAX_CONCURRENT` | `3` | Agent queue max workers |
+
+### Authelia Bypass
+
+API routes at `/api/*` bypass Authelia via `resources: ['^/api/']` in Authelia `access_control.yaml`. Frontend (`/`) requires SSO login.
+
+## API Endpoints
+
+| Endpoint | Method | Auth | Description |
+|----------|--------|------|-------------|
+| `/api/health` | GET | None | Health check |
+| `/api/config` | GET | None | Auth config (token required?) |
+| `/api/videos` | GET | None | List videos (status, channel, limit params) |
+| `/api/videos/fetch` | POST | Bearer | Trigger subscription sync |
+| `/api/videos/{id}/summarize` | POST | Bearer | Trigger single video summary |
+| `/api/videos/{id}/status` | POST | Bearer | Update status (pending/read/skipped) |
+| `/api/notes` | GET | None | List all notes |
+| `/api/notes/{id}` | GET/POST | Bearer (POST) | Get/save note |
+| `/api/notes/migrate` | POST | Bearer | Migrate localStorage notes to server |
+| `/api/agent-queue` | GET/POST | Bearer (POST) | List/add agent queue items |
+| `/api/channels` | GET | None | List distinct channels |
+| `/api/auto_categories.json` | GET | None | Channel category mapping |
+
+## Sprint Plan
+
+### Sprint 1 ✅ (Current — Deployed)
+- Agent queue worker (disabled by default)
+- Proxy support for fetcher + summarizer
+- Auto-mark-read on navigate away (not timer)
+- VPS fetch + summarize cron
+- run_tests.sh fixes
+- All 2,674 videos summarized (title-based)
+
+### Sprint 2 — UX Overhaul (Next)
+- Full-text search across video titles
+- Batch select → mark read/skip
+- Infinite scroll / pagination
+- Display duration + view count
+- Transcript accordion in detail panel
+
+### Sprint 3 — Polish & Monitoring
+- Stats dashboard (videos/day, channel health, sync lag)
+- Auto-purge old read videos (configurable TTL)
+- Broken channel detection (404 on RSS)
+- Cron failure alerts via Telegram bot
+- Scheduled DB backup to MacBook
+
+### Sprint 4 — Code Quality
+- Split `main.py` into routers
+- Cursor/offset pagination on API
+- Rate limiting middleware
+- aiosqlite connection pooling
+
+## Known Issues
+
+### Transcripts Blocked
+YouTube transcripts inaccessible from all our locations:
+- **Oracle Cloud VPS**: YouTube blocks cloud provider IPs
+- **MacBook (China)**: GFW blocks YouTube API endpoints
+- **YouTube Data API captions.download**: 403 Forbidden for non-owned videos
+
+**Workaround**: Title + description-based summaries via Qwen. Full transcripts need a Japanese residential proxy ($3-5/month).
+
+### auth.jimmygan.com from LAN
+Accessing `auth.jimmygan.com` from LAN shows Immich (photos.jimmygan.com) instead of Authelia. Caused by NAS port 443 routing — DSM nginx still owns port 443 and routes Immich for unmatched domains.
+
+### Gitea SSL
+Uses Let's Encrypt **staging** certificates (acme-staging) — browsers may show untrusted cert warnings.
+
+## Development
+
+### Local Setup
+```bash
+cd ~/repos/t-youtube
+# Backend
+cd backend && python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt
+python3 main.py  # runs on :8000
+
+# Frontend
+cd frontend && npm install && npm run dev  # runs on :5173 with API proxy
+```
+
+### Testing
+```bash
+cd ~/repos/t-youtube && bash run_tests.sh
+# 23 tests across db, fetcher, summarizer, main, agent_worker
+```
+
+### Deploy to NAS
+```bash
+cd ~/repos/t-youtube
+git add -A && git commit -m "..." && git push gitea main:master
+ssh nas 'cd /volume1/docker/t-youtube && git pull origin master && docker compose up -d --build'
+```

frontend/src/routes/YoutubeDashboard.svelte

@@ -22,10 +22,36 @@
   let editingCategories = $state(false);
   let prevSelectedId = $state(null);
   
+  // Search state
+  let searchQuery = $state('');
+  let searchResults = $state([]);
+  let isSearching = $state(false);
+  let showSearchResults = $state(false);
+  
+  // Batch select state
+  let selectedVideoIds = $state(new Set());
+  let showBatchBar = $state(false);
+  
+  // Stats state
+  let statsData = $state(null);
+  let showStats = $state(false);
+  
   // Agent queue state
   let agentItems = $state([]);
   let agentLoading = $state(false);
   
+  // Stats fetcher
+  async function loadStats() {
+    try {
+      const res = await fetch('/api/stats');
+      if (res.ok) statsData = await res.json();
+    } catch(e) {}
+  }
+  
+  $effect(() => {
+    if (activeTab === 'stats' && !statsData) loadStats();
+  });
+  
   $effect(() => {
     if (activeTab === 'pending' && channelFilter) {
       loadVideos();
@@ -51,6 +77,83 @@
   
   let noteIds = $derived(new Set(Object.keys(notes).filter(k => notes[k] && notes[k].trim())));
   
+  // Search debouncer
+  let searchTimeout = null;
+  async function handleSearch(query) {
+    searchQuery = query;
+    if (!query || query.length < 2) {
+      showSearchResults = false;
+      return;
+    }
+    isSearching = true;
+    try {
+      const res = await fetch(`/api/videos/search?q=${encodeURIComponent(query)}&limit=50`);
+      if (res.ok) {
+        searchResults = await res.json();
+        showSearchResults = true;
+      }
+    } catch(e) {
+      console.error('Search failed:', e);
+    } finally {
+      isSearching = false;
+    }
+  }
+  
+  // Batch operations
+  function toggleSelect(videoId) {
+    const newSet = new Set(selectedVideoIds);
+    if (newSet.has(videoId)) {
+      newSet.delete(videoId);
+    } else {
+      newSet.add(videoId);
+    }
+    selectedVideoIds = newSet;
+    showBatchBar = newSet.size > 0;
+  }
+  
+  function selectAll() {
+    selectedVideoIds = new Set(filteredVideos.map(v => v.video_id));
+    showBatchBar = true;
+  }
+  
+  function clearSelection() {
+    selectedVideoIds = new Set();
+    showBatchBar = false;
+  }
+  
+  async function batchUpdate(status) {
+    if (selectedVideoIds.size === 0) return;
+    const ids = Array.from(selectedVideoIds);
+    try {
+      const res = await fetch('/api/videos/batch', {
+        method: 'POST',
+        headers: getAuthHeaders({ 'Content-Type': 'application/json' }),
+        body: JSON.stringify({ video_ids: ids, status })
+      });
+      if (res.ok) {
+        // Remove selected from local list
+        videos = videos.filter(v => !ids.includes(v.video_id));
+        clearSelection();
+        loadPendingCount();
+      }
+    } catch(e) {
+      console.error('Batch update failed:', e);
+    }
+  }
+  
+  function formatDuration(seconds) {
+    if (!seconds || seconds < 60) return '';
+    const m = Math.floor(seconds / 60);
+    const s = seconds % 60;
+    return `${m}:${s.toString().padStart(2, '0')}`;
+  }
+  
+  function formatViews(count) {
+    if (!count || count < 1000) return count || '';
+    if (count < 1000000) return `${(count / 1000).toFixed(1)}K`;
+    return `${(count / 1000000).toFixed(1)}M`;
+  }
+  
   function toggleTheme() {
     theme = theme === 'dark' ? 'light' : 'dark';
     localStorage.setItem('theme', theme);
@@ -305,7 +408,9 @@
   function handleKeydown(e) {
     if (!videos.length) return;
     if (e.target.tagName === 'TEXTAREA' || e.target.tagName === 'INPUT') return;
-    const idx = filteredVideos.findIndex(v => v.video_id === selectedVideo?.video_id);
+    // If no video selected, let keyboard scroll the page naturally
+    if (!selectedVideo) return;
+    const idx = filteredVideos.findIndex(v => v.video_id === selectedVideo.video_id);
     let nextIdx = idx;
     if (e.key === 'j' || e.key === 'ArrowDown') {
       e.preventDefault();
@@ -374,6 +479,13 @@
     if (!d) return '';
     return d.slice(0, 10);
   }
+  
+  // Stats display helpers
+  function formatStatsNumber(n) {
+    if (n >= 1000000) return `${(n/1000000).toFixed(1)}M`;
+    if (n >= 1000) return `${(n/1000).toFixed(1)}K`;
+    return n.toString();
+  }
 </script>
 
 <div style="background:var(--bg-primary);color:var(--text-primary);height:100vh;overflow:hidden;display:flex;flex-direction:column">
@@ -429,15 +541,34 @@
     <aside style="width:280px;background:var(--bg-secondary);border-right:1px solid var(--border)" class="flex flex-col flex-shrink-0">
       <!-- Tabs -->
       <div class="flex p-1.5" style="border-bottom:1px solid var(--border)">
-        {#each ['pending','read','skipped','notes','🤖 Agent'] as tab}
+        {#each ['pending','read','skipped','notes','stats','🤖 Agent'] as tab}
           <button onclick={() => activeTab = tab}
             class="flex-1 py-1.5 text-center text-[11px] font-bold rounded uppercase transition"
             style="color: {activeTab === tab ? 'var(--accent)' : 'var(--text-muted)'}">
-            {tab === 'pending' && pendingCount ? `Pending (${pendingCount})` : tab}
+            {tab === 'pending' && pendingCount ? `Pending (${pendingCount})` : tab === 'stats' ? '📊 Stats' : tab === '🤖 Agent' ? '🤖 Agent' : tab}
           </button>
         {/each}
       </div>
 
+      <!-- Search bar (shown in pending/read/skipped tabs) -->
+      {#if activeTab !== '🤖 Agent' && activeTab !== 'stats' && activeTab !== 'notes'}
+        <div class="px-2 py-1.5" style="border-bottom:1px solid var(--border)">
+          <div class="relative">
+            <input type="text" 
+                   bind:value={searchQuery}
+                   oninput={handleSearch}
+                   placeholder="Search videos..."
+                   class="w-full text-[11px] p-1.5 rounded border"
+                   style="background:var(--bg-primary);color:var(--text-primary);border-color:var(--border)"
+                   onfocus={() => { if (searchResults.length) showSearchResults = true; }}
+                   onblur={() => setTimeout(() => showSearchResults = false, 200)}>
+            {#if isSearching}
+              <div class="absolute right-2 top-1.5 text-[9px] animate-spin" style="color:var(--accent)">↻</div>
+            {/if}
+          </div>
+        </div>
+      {/if}
+
       <!-- Agent Queue tab content -->
       {#if activeTab === '🤖 Agent'}
         <div class="flex-1 overflow-y-auto" style="border-color:var(--border)">
@@ -475,6 +606,55 @@
             {/each}
           {/if}
         </div>
+      {:else if activeTab === 'stats'}
+        <!-- Stats Dashboard -->
+        <div class="flex-1 overflow-y-auto p-3 space-y-3">
+          {#if !statsData}
+            <div class="flex items-center justify-center py-8 text-xs" style="color:var(--text-muted)">Loading stats...</div>
+          {:else}
+            <!-- Summary cards -->
+            <div class="grid grid-cols-2 gap-2">
+              <div class="p-2 rounded border text-center" style="background:var(--bg-tertiary);border-color:var(--border)">
+                <div class="text-lg font-black" style="color:var(--accent)">{formatStatsNumber(statsData.total || 0)}</div>
+                <div class="text-[9px]" style="color:var(--text-muted)">Total Videos</div>
+              </div>
+              <div class="p-2 rounded border text-center" style="background:var(--bg-tertiary);border-color:var(--border)">
+                <div class="text-lg font-black" style="color:var(--accent)">{statsData.summary_rate || 0}%</div>
+                <div class="text-[9px]" style="color:var(--text-muted)">Summary Rate</div>
+              </div>
+              <div class="p-2 rounded border text-center" style="background:var(--bg-tertiary);border-color:var(--border)">
+                <div class="text-lg font-black" style="color:var(--accent)">{statsData.channels || 0}</div>
+                <div class="text-[9px]" style="color:var(--text-muted)">Channels</div>
+              </div>
+              <div class="p-2 rounded border text-center" style="background:var(--bg-tertiary);border-color:var(--border)">
+                <div class="text-lg font-black" style="color:var(--accent)">{statsData.last_7_days || 0}</div>
+                <div class="text-[9px]" style="color:var(--text-muted)">Last 7 Days</div>
+              </div>
+            </div>
+            
+            <!-- Status breakdown -->
+            <div class="p-2 rounded border" style="border-color:var(--border)">
+              <div class="text-[10px] font-bold mb-2" style="color:var(--text-muted)">STATUS BREAKDOWN</div>
+              {#each Object.entries(statsData.status_breakdown || {}) as [status, count]}
+                <div class="flex justify-between items-center py-1">
+                  <span class="text-[11px]" style="color:var(--text-primary)">{status}</span>
+                  <span class="text-[11px] font-bold" style="color:var(--accent)">{count}</span>
+                </div>
+              {/each}
+            </div>
+            
+            <!-- Top channels -->
+            <div class="p-2 rounded border" style="border-color:var(--border)">
+              <div class="text-[10px] font-bold mb-2" style="color:var(--text-muted)">TOP CHANNELS</div>
+              {#each (statsData.top_channels || []).slice(0, 8) as ch}
+                <div class="flex justify-between items-center py-1">
+                  <span class="text-[10px] truncate flex-1 mr-2" style="color:var(--text-primary)">{ch.name}</span>
+                  <span class="text-[10px]" style="color:var(--text-muted)">{ch.count}</span>
+                </div>
+              {/each}
+            </div>
+          {/if}
+        </div>
       {:else}
         <!-- Channel filter bar -->
         {#if (allChannels.length || channels.length) > 1}
@@ -541,14 +721,31 @@
             {@const i = filteredVideos.indexOf(video)}
             <button onclick={() => { selectedVideo = video; }}
               data-video-index={i}
-              class="w-full text-left p-3 transition flex items-start space-x-3 outline-none focus:outline-none"
+              class="w-full text-left p-2.5 transition flex items-start space-x-2 outline-none focus:outline-none"
               style="border-left:2px solid {selectedVideo?.video_id === video.video_id ? 'var(--accent)' : 'transparent'};
                      background:{selectedVideo?.video_id === video.video_id ? 'var(--bg-tertiary)' : 'transparent'}">
-              <img src={video.thumbnail_url} alt="" class="w-20 h-12 rounded object-cover flex-shrink-0" style="background:var(--bg-tertiary)" />
+              <!-- Batch select checkbox -->
-              <div class="min-w-0">
+              <input type="checkbox" 
+                     checked={selectedVideoIds.has(video.video_id)}
+                     onclick={(e) => { e.stopPropagation(); toggleSelect(video.video_id); }}
+                     class="w-3 h-3 mt-1 rounded flex-shrink-0"
+                     style="accent-color:var(--accent)">
+              <img src={video.thumbnail_url} alt="" class="w-16 h-10 rounded object-cover flex-shrink-0" style="background:var(--bg-tertiary)" />
+              <div class="min-w-0 flex-1">
                 <span class="block text-[10px] font-semibold truncate" style="color:var(--accent)">{video.channel_name}</span>
-                <h3 class="text-xs font-bold line-clamp-2 leading-snug" style="color:var(--text-primary)">{video.title}</h3>
+                <h3 class="text-[11px] font-bold line-clamp-2 leading-snug" style="color:var(--text-primary)">{video.title}</h3>
-                <span class="text-[10px] mt-0.5 block" style="color:var(--text-muted)">{video.publish_date || ''} {noteIds.has(video.video_id) ? '📝' : ''}</span>
+                <div class="flex items-center gap-2 mt-0.5">
+                  <span class="text-[9px]" style="color:var(--text-muted)">{video.publish_date || ''}</span>
+                  {#if video.duration}
+                    <span class="text-[9px]" style="color:var(--text-muted)">{formatDuration(video.duration)}</span>
+                  {/if}
+                  {#if video.view_count && video.view_count > 0}
+                    <span class="text-[9px]" style="color:var(--text-muted)">{formatViews(video.view_count)}</span>
+                  {/if}
+                  {#if noteIds.has(video.video_id)}
+                    <span class="text-[9px]">📝</span>
+                  {/if}
+                </div>
               </div>
             </button>
           {/each}
@@ -556,6 +753,17 @@
       {/if}
     </aside>
 
+    <!-- Batch action bar -->
+    {#if showBatchBar}
+      <div class="fixed bottom-4 left-1/2 transform -translate-x-1/2 z-40 px-4 py-2 rounded-full shadow-lg flex items-center space-x-3"
+           style="background:var(--bg-secondary);border:1px solid var(--border)">
+        <span class="text-xs font-bold" style="color:var(--accent)">{selectedVideoIds.size} selected</span>
+        <button onclick={() => batchUpdate('read')} class="text-[10px] px-3 py-1 text-white font-bold rounded" style="background:var(--accent)">Read All</button>
+        <button onclick={() => batchUpdate('skipped')} class="text-[10px] px-3 py-1 font-bold rounded border" style="color:var(--text-secondary);border-color:var(--border)">Skip All</button>
+        <button onclick={clearSelection} class="text-[10px] px-3 py-1 font-bold rounded border" style="color:var(--text-muted);border-color:var(--border)">Clear</button>
+      </div>
+    {/if}
+
     <!-- Right: summary panel -->
     <main bind:this={summaryPanel} onscroll={handleSummaryScroll} id="summary-panel" class="flex-1 overflow-y-auto" style="background:var(--bg-primary);min-height:0">
       {#if activeTab === '🤖 Agent'}
@@ -606,6 +814,14 @@
                 onclick={() => { channelFilter = selectedVideo.channel_name; }}
                 title="Filter by this channel">{selectedVideo.channel_name}</span>
               <h1 class="text-lg font-bold leading-snug" style="color:var(--text-primary)">{selectedVideo.title}</h1>
+              <div class="text-[10px]" style="color:var(--text-muted)">{selectedVideo.publish_date || ''} 
+                {#if selectedVideo.duration}
+                  • {formatDuration(selectedVideo.duration)}
+                {/if}
+                {#if selectedVideo.view_count}
+                  • {formatViews(selectedVideo.view_count)} views
+                {/if}
+              </div>
               <div class="flex space-x-2">
                 <button onclick={() => updateStatus(selectedVideo.video_id, 'read')} class="text-[10px] px-2.5 py-1 text-white font-bold rounded transition" style="background:var(--accent)">Read (r)</button>
                 <button onclick={() => updateStatus(selectedVideo.video_id, 'skipped')} class="text-[10px] px-2.5 py-1 font-bold rounded border transition" style="color:var(--text-secondary);border-color:var(--border)">Skip (s)</button>
@@ -622,7 +838,8 @@
             <div class="space-y-1.5">
               <textarea id="note-textarea" rows="3" class="w-full text-xs p-2 rounded border resize-none"
                 style="background:var(--bg-tertiary);color:var(--text-primary);border-color:var(--border)"
-                placeholder="Write instructions for the agent here...&#10;e.g. 'Add this gyroid lattice workflow as a Hermes skill'"
+                placeholder="Write instructions for the agent here...
+e.g. 'Add this gyroid lattice workflow as a Hermes skill'"
                 value={notes[selectedVideo.video_id] || ''}
                 oninput={e => notes[selectedVideo.video_id] = e.target.value}
                 use:autoFocus

run_sync.py

@@ -0,0 +1,61 @@
+#!/usr/bin/env python3
+import asyncio, os, sys, logging
+
+REPO_ROOT = os.path.dirname(os.path.abspath(__file__))
+BACKEND = os.path.join(REPO_ROOT, "backend")
+ENV_PATH = os.path.join(REPO_ROOT, ".env")
+if os.path.exists(ENV_PATH):
+    with open(ENV_PATH) as f:
+        for line in f:
+            line = line.strip()
+            if line and not line.startswith("#") and "=" in line:
+                key, _, value = line.partition("=")
+                value = value.strip().strip("'\"")
+                os.environ[key] = value
+
+sys.path.insert(0, BACKEND)
+os.chdir(BACKEND)
+
+from db import init_db, DB_PATH
+from fetcher import fetch_subscriptions
+from summarizer import summarize_video
+import aiosqlite
+
+BATCH_SIZE = int(os.environ.get("SUMMARIZE_BATCH", "20"))
+BATCH_DELAY = float(os.environ.get("SUMMARIZE_BATCH_DELAY", "2.0"))
+
+logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
+
+async def main():
+    await init_db()
+    
+    count = await fetch_subscriptions()
+    print(f"Fetched {count} new videos")
+    
+    async with aiosqlite.connect(DB_PATH) as db:
+        cur = await db.execute("SELECT COUNT(*) FROM videos WHERE summary IS NULL OR summary = ''")
+        total = (await cur.fetchone())[0]
+    
+    if total == 0:
+        print("No pending videos to summarize.")
+        return
+    
+    print(f"Processing batch of {BATCH_SIZE} (of {total} pending)...")
+    
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        cur = await db.execute("SELECT video_id FROM videos WHERE summary IS NULL OR summary = '' LIMIT ?", (BATCH_SIZE,))
+        rows = await cur.fetchall()
+    
+    done = 0
+    for row in rows:
+        await summarize_video(row["video_id"])
+        done += 1
+        if done < len(rows):
+            await asyncio.sleep(BATCH_DELAY)
+    
+    remaining = total - done
+    print(f"Batch done: {done}/{len(rows)}, {remaining} pending")
+
+if __name__ == "__main__":
+    asyncio.run(main())