PlotVeil
Shh, keep the surprises.
Comments arrive covered. Anything that would spoil what you are watching stays that way until you decide otherwise.
No account · No analytics · YouTube only · Open about what it sends
NewJudgments run on Jev, TypeSafe's System One model. 70–500 ms, no generated text
Harrow Bay: Episode 6 breakdown
live demo
Covered first. Judged second.
Most blockers match keywords and lose the race against your eyes. PlotVeil inverts the order: nothing is readable until it has been checked, and a failed check keeps the cover on.
Covered on arrival
As comments load, PlotVeil covers them in place. The original text is hidden rather than blurred, and the layout does not jump. Likes, replies and translation rows stay where they were.
Read for meaning
Each comment is judged against the video title, channel, and any other works you asked to protect. No keyword list, no "death" and "dies" blocklist. An oblique hint in Japanese counts as much as an explicit one in English.
Reveal, and re-hide
Clean comments come back instantly. Spoilers stay covered until you choose Show anyway. The Veil Keeper stays beside the comment, so one click puts the cover back.
Jev answers in probabilities.
PlotVeil runs on Jev, TypeSafe AI's System One model, released September 2026. Jev does not write text. It takes unstructured state in and returns typed, calibrated probabilities out, which is the shape of the question PlotVeil needs answered: would this comment ruin the episode?
{ "model": "jev-latest", "state": { "video_title": "Harrow Bay: Episode 6 breakdown", "channel": "Late Night Rewatch", "comments": { "c42": "Book readers: they kept…" } }, "questions": { "c42": { "type": "noul", "instructions": "state.comments.c42 reveals a plot point of the work being watched" } } }
{ "answers": { "c42": { "type": "noul", "noul": 0.93 } } }
TypeSafe's published end-to-end range for Jev, 40–200× faster than frontier LLMs. Fast enough that a comment can be covered, checked and uncovered before you scroll past it.
Jev samples every answer in parallel, so twenty comments are one question set, not twenty round trips.
No free text means no hallucinated verdict and no type errors to parse around. A comment cannot talk the model into letting it through.
Every answer is a number, not a yes or no. The sensitivity setting moves the cut-off on that number, so changing it changes what stays covered.
Changing sensitivity re-uses the probabilities already in hand: no second request, no second charge. Requests go through PlotVeil's own Cloudflare Worker, which holds the API key and strips your installation ID before the model ever sees the text. Latency figures above are TypeSafe's; our own round trip measured around 0.7s in testing, and your network will differ.
What the Keeper is telling you.
No badge counts, no toasts, no pop-ups mid-video. The Keeper in your toolbar, on the player and beside each covered comment shows what protection is doing right now. Hover or click it for the details.
Neutral idle
Nothing to do here: not a video page, or the comments have not loaded yet. It will not call this "safe."
Scout checking
Eyes open, first batch in flight. Everything waiting on a verdict is already covered.
Calm clear so far
Nothing spoiler-ish in what has been checked. Honest about the "so far": comments still loading are still pending.
Guard protecting
Spoilers found and held under cover. This is the state you will see most on an active comment section.
Peek revealed
You chose to read something. The Keeper waits beside it, one click from putting the cover back.
Concerned error
A check timed out, the quota ran out, or a comment was too long. Covers stay on, the reason is spelled out, and nothing flashes red at you.
Closed off
Protection paused, or you are on a channel you marked as trusted. Covers come off, nothing is sent, nothing is judged.
Quiet by design
Comment covers are one line of plain text. The player entry only opens a panel when you click it. Under 520px, it gets out of the way entirely.
What it does, in full.
- Covers before it judges. Text is hidden with
visibility, not blur, so there is no half-readable moment. - Protects works you are not even watching. Add titles you are behind on, and comments that spoil them get covered here too.
- Survives the page. The covers keep up through infinite scroll, replies, edited comments, theatre mode, SPA navigation, light and dark.
- Trusted channels. Somewhere you know is spoiler-free? Turn protection off for that channel only.
- Reversible both ways. One click lifts a cover, one click puts it back.
What it will not claim.
- Comments that have not loaded yet have not been checked. "Clear so far" never means "all clear."
- A judgment can be wrong in both directions. That is why every cover lifts with one click and every reveal goes back.
- Video titles, thumbnails and recommendations are untouched for now. This is a comment tool.
- YouTube comments only today. Bilibili comments and danmaku share the same judgment core, but they are not supported yet and are not sold as if they were.
To judge a comment, PlotVeil has to send it. Here is everything that goes out.
While protection is on, comments near your viewport go out with the video title, the channel, any works you asked to protect, and a random installation ID used only for the daily quota. The model never receives that ID.
Never sent
Your Google account. Cookies. Watch history. Page URLs. Comment author names. Anything from any site other than youtube.com.
Never present
Analytics, tracking pixels, third-party scripts and ad identifiers are absent from both the extension and this page. The fonts you are reading are served from this domain.
Before you install.
Does it slow the comment section down?
Comments appear as fast as YouTube renders them, with the cover already on. Jev's speed is the reason that cover can lift again within a fraction of a second rather than after a visible pause. Batches of up to 20 comments go out at once, with at most two requests in flight.
What happens when a check fails?
The cover stays on. A failed, timed-out or over-quota check is never treated as "safe". The Keeper turns Concerned, tells you which of those it was, and offers a retry when retrying would help. You can still reveal any individual comment by hand.
Will it hide comments that are not spoilers?
Sometimes, especially on Cautious. Every cover is one click from lifting, and lifting one never asks you to disable protection for the rest.
Is it free?
Yes, with a daily judgment quota per installation to keep the model costs sane. If you hit it, PlotVeil says so plainly and tells you when it resets.
Does it work on other browsers or sites?
Chrome and Chromium-based browsers, on youtube.com. Firefox and the next set of sites are on the roadmap; neither is claimed as working today.
Who makes it?
One independent developer. Not affiliated with YouTube, Google or TypeSafe AI.
Go read the comments.
The good ones are still down there. PlotVeil keeps the rest covered until you have finished the episode.
