Big Idea

The argument in one line.

Routing video-editing prompts through Jev, a non-reasoning tool-selection model, turns editing into instant, free tool calls instead of slow LLM reasoning, so one person can source, cut, caption, and publish without touching a timeline by hand.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You maintain an editing pipeline, manual or AI-assisted, and want a working reference for wiring a fast decision-model in front of it instead of a slow reasoning LLM.
You're evaluating tool-routing patterns for agents and want a concrete example of a non-reasoning choice/score/yes-no model dispatching real work.
You use Claude Code plus open-source tools like FFmpeg, Whisper, Remotion, or Obsidian for content production and want to see them chained together.

SKIP IF…

You want a finished video editor to download and use as-is — HyperEdit is the creator's own open-source project, not a polished product.
You're looking for editing technique or storytelling advice — this is a tools-and-architecture video, not a craft tutorial.

TL;DR

The full version, fast.

Kev connects Jev, a model from TypeSafe AI that resolves every prompt into one of three answer types instead of open-ended reasoning, to his open-source video editor HyperEdit. He demos plain-English prompts triggering real edits: extracting audio, generating yellow captions, and stripping dead air, then shows an Obsidian vault letting Jev pull matching logos and clips onto the timeline by keyword search. He frames this as solving the tradeoff between slow open-source models and token-expensive frontier models, since Jev's tool calls cost nothing, and pitches the endpoint as a pipeline that edits, produces, and posts to social with no human touching the timeline.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:51

01 · Cold open + HyperEdit recap

Kev reintroduces HyperEdit, his open-source AI video editor, and teases the Jev AI integration as the biggest update yet.

00:51 – 01:29

02 · Live demo: extract audio, auto-caption, remove dead air

Plain-English prompts typed into HyperEdit's chat panel execute as real edits: audio extraction, yellow captions, and a dead-air cut that shrinks the timeline from 2:30 to 2:28.

01:29 – 02:27

03 · Obsidian agent pulls media straight onto the timeline

A connected Obsidian vault lets Jev search by keyword ('Claude', 'subway surfer') and drop matching logos and a video clip directly onto the timeline.

02:27 – 02:57

04 · Why this is the endgame: full automation

Kev frames the real payoff: with Jev wired in, an agent can go from source to cut to caption without a human at the timeline at all.

02:57 – 05:16

05 · What Jev actually is, and how it routes every prompt

Jev resolves prompts via three answer types (choice, score, yes/no) fired at once instead of reasoning, and Kev diagrams the 'conductor' architecture routing calls to FFmpeg, Whisper, Remotion, or Claude, closing with the setup ask: grab the codebase and a Jev API key.

Atomic Insights

Lines worth screenshotting.

Jev is described as a model that doesn't reason: it answers every prompt with one of three response types (a pick from up to 255 options, a numeric score, or yes/no), fired simultaneously instead of step-by-step thinking.
Because Jev skips reasoning entirely, it's pitched as able to dispatch a video-editing tool call instantly and without spending LLM tokens, versus routing every edit through a full reasoning model.
In the demo, a single plain-English prompt ('remove dead air') cut a 2:30 video down to 2:28 automatically.
The editor's chat panel escalates to Claude only when Jev's three answer types can't resolve a request, keeping Claude calls, and their token cost, to the exception case rather than the default path.
A connected Obsidian vault lets a text search ('Claude', 'subway surfer') pull matching logos and video clips straight onto the timeline, with no manual file browsing.
The architecture routes every prompt through five tools: FFmpeg twice (audio and video separately), Whisper for transcription, Remotion for animation, and Claude as the fallback, with Jev acting as the dispatcher at the center.
The stated tradeoff being solved: open-source/local models are slow, frontier models are expensive per token — a fast non-reasoning dispatcher is pitched as removing that choice for routine editing actions.
The end state described is a pipeline where an agent doesn't just cut the video — it also handles post-production and posts the finished clip to social media with no human touching the timeline.
HyperEdit, the editor being demoed, is the creator's own open-source project, released earlier and already forked by hundreds of users before this particular integration shipped.

Takeaway

Route routine work through a fast dispatcher

TOOL ROUTING

Splitting 'decide what to do' into three cheap answer types instead of open-ended reasoning turns routine tool calls into instant, free actions.

Constraining a dispatcher to a small set of answer types (pick one, score it, yes or no) removes the need for step-by-step reasoning on routine decisions, which is what makes each call fast and cheap.
Wiring a fast, non-reasoning layer in front of a slower, more expensive model lets the expensive model handle only the requests the cheap layer can't resolve, instead of being the default path for everything.
Connecting a searchable knowledge base, here an Obsidian vault, directly to a production tool lets a text query pull the right asset onto the timeline instead of manual file browsing.
Automating a single step, like removing dead air, is a small win; chaining source-to-cut-to-post into one pipeline is what actually removes the human from the loop.
An open-source project that others can fork becomes its own distribution channel: hundreds of forks happened before this particular integration even shipped.

Glossary

Terms worth knowing.

HyperEdit: The creator's open-source AI video editor, run locally, where prompts typed into a chat panel trigger real timeline edits.
Jev: A model from TypeSafe AI, described as answering only by picking from a fixed set of options, a score, or yes/no, rather than by open-ended reasoning, pitched as faster and token-free versus a reasoning LLM.
Obsidian: A note-taking app whose file vault is wired into the editor here as a searchable source of images and video clips.
Remotion: A tool for programmatically generating video animations, listed as one of the tools Jev can call.
token: A unit of text an AI model is billed per; the video's central claim is that Jev's own tool calls don't consume any.

Resources

Things they pointed at.

00:00toolHyperEdit (GitHub repo) ↗

00:00linkPrevious HyperEdit tutorial ↗

00:28toolJev (TypeSafe AI console) ↗

01:29toolObsidian ↗

04:51productCreator OS ↗

Quotables

Lines you could clip.

00:00

“Boy, oh boy, do I have an absolute treat for you guys today.”

high-energy cold-open hook→ TikTok hook↗ Tweet quote

02:57

“Jev is the first AI model that doesn't think. It just knows answers.”

the one-line pitch the whole video is built around→ IG reel cold open↗ Tweet quote

04:22

“we were kind of in between a rock and a hard place... and then Jev comes in and absolutely solves this entire problem”

names the tradeoff being sold in one breath→ newsletter pull-quote↗ Tweet quote

04:51

“the agent... doesn't just do the editing, it'll do post-production, it'll actually post the content to social media”

the full-automation payoff line→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphor

Boy, oh boy, do I have an absolute treat for you guys today. So on this channel, I built and released this open -sourced video editor called HyperEdit, which lets our AI agents literally handle an entire video edit and the timeline, which helps us replace tools like CapCut and Adobe Premiere Studio. And since releasing, hundreds of you have forked it and are using it in real time.

And I've released three video updates about different things that we've added to this project over time. And in this video, I have the biggest update yet because this new tool called JevAI literally completely transforms video editing workflows.

And in this video, I'm going to explain how Jeb AI literally is going to revolutionize the video editing and content creation game and how it works inside of our hyper edit code base, as well as some really cool examples. This is the one -stop shop for Jeb AI video editing. Let's get into it, baby.

But first I got to show you this demo because it's... absolutely insane all right so check this out on the left hand side we have the hyper edit project on the right hand side we have the jev knowledge source of all of these different pieces of media that we now have access inside of the video editor so now let me just show you how fast this thing actually works so first prompt we're going to do extract audio boom audio gets extracted next prompt we're going to do add captions it's going to ask us what color let's say we want it in yellow boom it's going to put that in there the video is currently 2 minutes and 30 seconds long and let's say we wanted to cut all of the dead air from the video which usually takes a lot of time as an editor we're just going to go remove dead air There we go.

So now the video is 2 minutes and 28 seconds. Now let's move on to the second feature that we have with Jev, which is the obsidian agent. So on the right hand side, we have all of these files and now I can source all of these files into workflows just by asking for it.

So let's test that out. Let's look for some Claude logos. So I'm going to type in Claude and it literally pulls in the Claude logo and imports it to our timeline, but it also has videos.

So let's see if we can find the Jev subway surfer video.

There we go. And it's importing that video as well to the timeline. It's been imported.

And now I can grab that three minute video and I can add it to the exact same timeline that I'm currently editing on. Now there is so much more to unpack and there's a bunch of other tutorials associated to hyperedit. But what you need to understand is that the goal of this is for you to see the different workflows that you can actually put together with AI.

And in that way, you can turn everything that you're working on in hyperedit into an automation because everything that I just showed you, for example, was just Jeff making a tool call. And all of that can be automated. We don't need to be here sitting here editing at all.

And that's where this creativity explosion is happening. That's why I even open sourced this video editor to begin with. Now guys, I want to explain what makes all of this possible.

So we're going to go through three quick chapters explaining what Jev is one more time really quickly. And then we're going to talk about how Jev actually works inside of that video editing workflow and how I'm using it to grow on socials. So first off, what is Jev?

So Jev is the first AI model that doesn't think. It just knows answers. by the tools that it's given.

And it gets these answers by asking three questions. So it can choose from 255 options, which you give it. It can score things on a rating that you give it as well.

And it can answer things yes or no. But it asks all of these questions at the same time. And because of this, it's able to tool call for video workflows at light speed, like I showed you guys, because there's no thinking, there's just action.

We're asking for a video to be cut. It'll go and find the dead air. We're asking for captions.

It'll go grab the caption tool. And so this saves tokens, but it also helps us work 100 times faster. Inside of the video editing workflow, here's how...

jev actually works so every single prompt that we send in is actually going to hit jev first and jev has a entire arsenal of options that it has to choose from based on what our prompt was and one of those options is relaying the message to cloud itself so if it's if if we ask it a question for the video editor and it doesn't have an answer it'll then route it to clot but in the examples that i showed you today it was able to source all of that information all of that media from the Obsidian tool that is attached to Jev.

It can create video animations with the Remotion tool that's attached. It's able to get transcriptions from the Whisper tool that runs locally on the computer. And FFmpeg is what is allowing Jev to make cuts in the video, extract audio, grab real word timestamps, and this.

all together allows jev to literally create end -to -end video pipelines without us spending any tokens so aside from it being really fast and really revolutionary it's also not costing us any tokens and that was the biggest bottleneck with the open source video editor because we were kind of in between a rock and a hard place because if you if you use open source models it's a lot slower if you use the cutting edge models it'll end up costing a lot of tokens and then jev comes in and absolutely solves this entire problem and if you guys are following me on my short form channels this is literally the formula that's allowing me to just create so much content at scale because once a video is created the agent and in this case jev doesn't just do the editing it'll do the editing it'll do post -production it'll actually post the content to social media now the exciting thing for you is all you need to do to set up hyper edit is to grab the code base and grab the jev api key and everything else is ready to go catch you next one

The Hook

The bait, then the rug-pull.

Kev already open-sourced an AI video editor, HyperEdit, that hundreds of people have forked. This video is about the piece he just bolted on: Jev, a model that doesn't reason so much as instantly pick an answer, now sitting in the driver's seat of the whole edit.

Frameworks

Named ideas worth stealing.

02:57list

Jev's three answer types

Choice (up to 255 options)
Score (rate this)
Yes/No

The three response shapes Jev is described as using to resolve every prompt simultaneously instead of reasoning step by step.

Steal fordesigning a fast dispatcher/router layer in front of any agent's tools

03:33concept

Jev as "the conductor"

Diagram showing Jev at the center, routing every prompt to FFmpeg (x2), Whisper, Remotion, or Claude, with Claude invoked only as a fallback.

Steal forexplaining a hub-and-spoke tool-routing architecture

CTA Breakdown

How they asked for the click.

VERBAL ASK

04:51link

“all you need to do to set up HyperEdit is to grab the code base and grab the Jev API key and everything else is ready to go”

Soft CTA folded into the sign-off rather than a hard pitch; points to the free GitHub repo and Jev API signup. His own Creator OS / Skool links only appear in the description, not spoken on camera.

MENTIONED ON CAMERA

00:00toolHyperEdit (GitHub repo) ↗

00:00linkPrevious HyperEdit tutorial ↗

00:28toolJev (TypeSafe AI console) ↗

01:29toolObsidian ↗

04:51productCreator OS ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.

OTHER LINKSAlso linked in the description.