🔍
Semantic matching
Not keywords — AI embeddings compare meaning. A video about ‘Transformers’ and one about ‘트랜스포머’ are recognized as the same topic automatically.
AI YouTube Filter
Filter out the noise,
keep what matters.
Paste any YouTube link and AI parses the transcript, semantically compares it against everything you've already watched, and tags it
🟢 Must watch / 🟡 Some overlap / 🔴 Skip.
3 free videos without signup · Bulk playlist support
Problem
The algorithm keeps recommending the same content with different titles
If you spend a lot of time on YouTube, this is familiar. Get into a topic, and the algorithm pushes endless similar videos. The catch — you have to watch the whole 10-minute video before realizing “wait, I've seen this before.” The wasted minutes pile up.
Tools like NotebookLM help, but they handle one video at a time, no bulk playlists, and no comparison against what you've already watched.
Solution
Not summarization. Automatic comparison.
📦
Bulk playlist
Paste a playlist URL instead of one link at a time. 30 videos sorted with a single click.
🎯
3-tier auto-classification
🟢 Must watch / 🟡 Some overlap / 🔴 Skip. Novelty % and the most-similar video are shown so you see exactly why.
How it works
25 seconds per video
- 1Transcript extraction~6s
Triple-redundant pipeline: youtube-transcript-api + proxy + yt-dlp fallback for stable extraction.
- 2Chunk + LLM analysis~10–15s
3,000-char chunks analyzed in parallel by Gemini 2.5 Flash Lite. Section titles, key concepts, and quotes extracted.
- 3Brief embedding~150ms
Title + tags + concepts + 3-line summary encoded as a 768-dim vector (Gemini embeddings).
- 4Library comparison + classification~50ms
pgvector RPC computes top-1 cosine similarity against your library → threshold-based grade.
Re-analyzing the same video takes 0.5s (global cache hit, zero API calls).
Who it's for
Built for
🤓
Heavy learners
If you follow fast-moving fields (AI, dev, design), every day. Compress watch time by surfacing only the videos with truly new information.
📚
Content curators
Newsletter writers, social posters, club organizers. Pick the densest video among many similar ones.
🎓
Course / research
Studying a single topic. Your library accumulates and TubeFilter automatically flags when the next video is rehashing what you've already covered.
Chrome Extension
Watch Later & channel home, in one go
On YouTube, hover any video card for [+ Add] or drag-select across multiple cards → one click sends them all to TubeFilter. No more copy-pasting playlist URLs.
Add to Chrome →Pricing
Simple subscriptions
Free
$0
180 min / month
- ✓3h analysis / month
- ✓Library search
- ✓Ads included
Basic
$2.99/ mo
1,500 min / month
- ✓25h analysis / month
- ✓Ad-free
- ✓Priority support
Plus
$6.99/ mo
3,600 min / month
- ✓60h analysis / month
- ✓All Basic features
- ✓Early access to new features
Billed in USD via Paddle. Full pricing →
FAQ
Frequently asked
Is TubeFilter free?
Yes — 30 minutes free without signup. After signing up, 180 minutes (3h) per month are free. Need more? Basic starts at $2.99/month.
What if a video has no captions?
We use auto-generated captions when available. Videos with captions disabled show as “transcript unavailable.”
How many videos in a playlist can I process at once?
Up to 20 new videos per request. Already-analyzed videos in your library are uncapped — total 100 per request.
Are other users' videos used for comparison?
No. Only videos in your own library count. User data is isolated via Postgres RLS.
How are payments handled?
Paddle (Merchant of Record). Cards, KakaoPay, NaverPay, Apple Pay supported. Card details never touch TubeFilter servers.
How do I cancel my subscription?
Cancel any time from the Account page. Service continues to the end of the current billing period, then ends automatically.
How is this different from NotebookLM?
NotebookLM deeply summarizes one source. TubeFilter is built to decide whether N videos are worth watching by comparing against your library — bulk processing + automatic dedup are the core difference.
Which AI models do you use?
Summarization: Google Gemini 2.5 Flash Lite. Embeddings: Gemini embedding-001 (768d). Semantic comparison: Postgres pgvector + cosine similarity.