Meet Otto, the editor that tells you when something is wrong

Otto is not a chatbot bolted to a sidebar. It reads the real state of your project — every clip, its media, its timing, its layout — and it runs the checks a second pair of eyes would run before you export.

Some of those checks are about things that will definitely ruin the video. A timeline with nothing at the start exports as black. A clip whose media is missing or whose upload never finished exports as a hole. Those are not opinions and Otto says them plainly.

Others are about intent, and those are the interesting ones. Otto samples actual frames from a clip to tell whether it is rendering black — not by checking one frame, which would be wrong constantly, but by sampling across the clip and requiring the evidence to agree. A voiceover that is silent, or near-silent, gets flagged. A clip that is pointing at a video file but is typed as an image gets flagged, because that combination produces a broken clip and a confused user. Footage held on screen longer than the footage actually lasts gets flagged, because the tail freezes.

And a smaller set are craft, not correctness: speech on screen with no captions, when most short-form video is watched muted. Long dead-air pauses. Music sitting at full volume over dialogue. Otto only raises those when you ask, or when you have clearly stalled.

An assistant that cries wolf gets switched off in ten seconds, and then it does not matter how good the rest of it is. Almost all of the design work went into restraint rather than capability.

Otto is silent by default. It only interrupts for things that are actually broken, never for style. It says each thing once and never repeats it. It will not speak while you are dragging a clip, typing, scrubbing or playing back — it waits for a gap. It has a hard ceiling of four unprompted remarks in a session, so even a genuinely messy project cannot turn into a lecture. And dismissing it is one click, with a difference between "not now" and "never" — where "never" means never, and survives a reload.

The false-positive problem got the same treatment. A title card over black is deliberate. A fade to black is deliberate. Footage graded almost to black is a look. Real night footage is just dark. Every one of those is, to a naive check, indistinguishable from a broken clip — so the check does not rely on brightness at all. It relies on structure: real footage carries grain, a gradient, compression texture. A frame that was never painted carries nothing. That single distinction is what makes the difference between an assistant that is useful and one that nags you about your own creative choices.

Otto listens. Click the mic, or hold the backtick key and say what you want — "what is wrong with this", "make it vertical", "cut the boring bits", "where is the problem". It answers out loud and writes down everything it says, so the whole thing works with the sound off.

Questions are answered on the spot and for free — how to export, what a tool does, what it thinks is wrong with your current edit, where the problem is on the timeline. Ask it to show you something and it will select the clip and move the playhead to it.

When you ask for an actual edit, Otto hands that to the same editing agent that powers Agent Mode, rather than doing its own thing. That is deliberate: it means a spoken edit gets the same billing, the same automatic refunds when a run fails, the same undo grouping, and the same visual check of its own work. One path, one set of guarantees.

Talking to Otto is free. Listening, answering questions, running its checks, pointing at problems, explaining what it found — none of that touches your AI credits, and it never will, because an assistant that bills you by the sentence cannot afford to be helpful.

Changing your video costs exactly what it costs in Agent Mode today, and Otto tells you the price before it spends anything. Nothing is ever spent by accident: a suggestion from Otto is an offer, and only a deliberate click on that offer starts a run. If Otto is not sure what you meant, it asks rather than guessing — an unclear request can never turn into a charge.

And when a run does not work, it says so. If it applied nothing, it tells you it applied nothing, because the alternative — claiming success it did not earn — is how people stop trusting a tool for good.

It is not a reviewer of taste. It will not tell you your cut is boring or your hook is weak, beyond a couple of well-worn craft prompts you have to ask for. Those judgements are yours and it would be obnoxious to pretend otherwise.

It is not watching you for our benefit. The checks run in your own browser against your own project. It is not a support macro, either — when it cannot fix something, like a file whose upload died, it says what you need to do instead of pretending to handle it.

It is on the desktop editor. Talking requires a browser with speech recognition, which today means Chrome or Edge; in Firefox Otto says so plainly and you type instead, and everything else works identically. And it can be turned off permanently in one click, from the character itself or the editor menu.

We could have shipped all of this as warning banners. We have banners. People do not read them, and a banner cannot be asked a follow-up question.

Otto is built from the Vidmoat logo — the rounded square, the gradient, the play triangle as its face — with arms and legs added, and it reacts to what is happening: it leans in when it is listening, hunches over when it is applying your edit, straightens up and points when it has found something. That is not decoration. When a tool is doing something on your behalf that takes several seconds, being able to tell at a glance whether it is listening, thinking or working is the difference between waiting and wondering.

It is also, we hope, likeable. The measure of whether that was worth doing is not whether people say it is cute. It is whether more people finish a video. We instrumented it for exactly that — how often Otto speaks, how often it is asked for help, how often a fix is accepted, how often someone reaches an export afterwards, and how often it is dismissed. If the dismissals climb and the exports do not, Otto is making things worse and we will say so.