HyperEdit: Open Source AI Video Editor With FFMPEG 2026

HyperEdit is an open source AI video editor that bundles FFMPEG, Remotion and Obsidian agents powered by Jev. You run it locally with npm install and npm run dev. This guide covers what the repo contains, how the FFMPEG and Remotion layers fit together, how it works as an MCP video editor with Claude or a coding agent, and where Creator OS fits once the edit is rendered.

  raw clip
     |
     v
  FFMPEG  ......... cut, trim, mux, encode
     |
     v
  Remotion  ....... animate the cut in React
     |
     v
  Obsidian agents . Jev reads the vault, plans edits
     |
     v
  rendered short  ->  Creator OS post

FFMPEG does the cutting, Remotion draws the frames, agents decide what to cut.

HyperEdit is Kevin’s own project, published at github.com/kevinbadi/hyperedit. The repository describes it as an AI-powered video editor with FFMPEG, Remotion and Obsidian agents baked in, powered by Jev. That phrase is the whole pitch, so this post unpacks it rather than adding claims the repo does not make. If you prefer to watch someone run it, the walkthroughs live on the KevBuildsApps YouTube channel.

What Is HyperEdit? The Open Source Video Editor With AI Agents Baked In

HyperEdit is a local app where the editing decisions can be handed to agents, but the actual rendering stays on your machine. That split is the part worth understanding.

Three pieces do three jobs:

FFMPEG handles the mechanical work. Cutting, trimming, muxing audio and video, encoding the output file. This is the layer that has to be fast and boring.
Remotion handles the visual layer. Remotion lets you write video as React components, so overlays, captions, counters and title cards become code you can reuse instead of timeline clips you drag by hand.
Obsidian agents handle the intent. You keep notes in an Obsidian vault, and agents read that context to plan what the edit should be. The project describes these agents as powered by Jev.

So the shape is: an agent proposes, FFMPEG executes, Remotion renders. You keep the source files. Nothing is uploaded to a render farm you do not control.

That matters if you already produce short form clips. A talking head recording becomes an edit plan in a vault, the plan becomes a cut, the cut becomes an MP4. The interesting question is not “can an AI edit video” but “where does the edit live” and in HyperEdit it lives in files on your disk.

Get started for $0

How HyperEdit Combines FFMPEG, Remotion and Obsidian Agents

Each layer has a clear boundary, which is what makes the combination workable rather than a pile of tools in one folder.

Layer	Job	What you touch
FFMPEG	Cut, trim, mux, encode	Command line, invoked by the app
Remotion	Animated overlays and frames as React	React components in the project
Obsidian agents	Read your vault, plan the edit	Markdown notes in a vault
Jev	The engine the agents run on	Configured in the app

A worked example makes it concrete. Say you record a six minute explanation of a pricing change. In your Obsidian vault you have a note for that topic with the key numbers and the three objections customers raise. An agent reads that note, decides the clip should open on the objection, and produces an edit plan. FFMPEG cuts the source video at the marks the plan gives it. Remotion lays the price on screen as an animated counter instead of a static graphic.

You get a short that reflects your own notes, not a generic template. That is the difference between a video editor with an AI button and an editor where the AI has your context.

For a deeper look at building animated talking head clips with skills and ffmpeg, read running AI video generation locally with coding agents.

HyperEdit vs Other Open Source Video Editors in 2026

The honest comparison is about shape, not winners. HyperEdit is an agent driven local app. Most open source editors are timeline editors you drive by hand.

If you want to drag clips around a GUI, a traditional open source non linear editor is the right tool and HyperEdit is not trying to replace it. If you want the edit to be described in text, planned by an agent and rendered from code, HyperEdit’s FFMPEG plus Remotion plus Obsidian stack is built for that.

Timeline editors put a human in front of a canvas. Fast for one off edits, slow to repeat across fifty clips.
HyperEdit puts the instructions in a vault and the rendering in code. Faster to repeat, and every edit is reproducible because the plan is a file.
Scripted ffmpeg pipelines repeat well but have no planning layer. HyperEdit adds the agents on top of the same idea.

One more difference worth naming: HyperEdit is open source and runs locally. Your footage stays on your disk. That is a property of the design, not a claim about anyone else’s product.

Can HyperEdit Work as an MCP Video Editor for Claude and Coding Agents?

That is the natural way to use it. The agents are already in the design, so a coding agent can drive the workflow the same way it drives any other local project.

The practical loop is ordinary agent work. You open the HyperEdit project in Claude Code or another coding agent, point it at the repo, and describe the edit you want. The agent reads the project, adjusts Remotion components, and runs the local commands. Because Remotion is React, the agent is editing code it can read. Because FFMPEG is a command line tool, the agent can run it and check the output.

Your Obsidian vault supplies the intent. The agent supplies the execution. You review the rendered file, not a wall of timeline clicks.

Creator OS takes the same idea and applies it to publishing. Creator OS ships a hosted MCP server and a CLI, so an AI agent can post, schedule, reply and read analytics on your accounts. Same pattern: your agent does the work, the tool exposes the actions. You can see the publishing side in the docs at creatoros.ca/docs/mcp.

How to Run HyperEdit Locally: npm Install, npm Run Dev and Your First Edit

Two commands get the dev server running. The repository lists exactly these:

npm install
npm run dev

That is the install path the README gives, so do not expect a CLI flag list beyond it. The repo also notes that the app was created using getmocha.com and links a Discord community for help, which is the place to go when the dev server does not come up on your machine.

A sane first session looks like this:

Clone github.com/kevinbadi/hyperedit and run the two commands above.
Open the dev server in your browser and confirm the editor loads before you bring in footage.
Put one short clip in the project and make a single cut with FFMPEG. Prove the pipeline works end to end.
Add one Remotion component, such as a caption or a counter, and render again.
Only then connect your Obsidian vault and let an agent plan a real edit.

Small steps matter here because three layers fail in three different ways. A broken FFMPEG path and a broken React component look nothing alike, and debugging one at a time is far quicker.

What You Can Build With an AI Video Editor Powered by Jev

The realistic output is a repeatable short form pipeline, not a one off showpiece. Here is what that looks like in practice.

An explainer series. One vault note per topic, each with the numbers and the objections. The agent turns each note into a cut plan. Remotion adds the same animated counter component every time, so episode twelve matches episode one. You record once a week and render five shorts.

A change log video. You keep product notes in Obsidian already. Point the agent at this week’s notes and get a 45 second update with the version number animated on screen. Push it to your channels as a post, not a hand edited file.

A course teaser. Pull the strongest 30 seconds from a long lesson, overlay the module title with Remotion, and use the same FFMPEG cut for three aspect ratios.

Notice what is missing from all three: no timeline, no manual keyframing, no re-export by hand. The work moves into notes and components you can edit next month.

For a related approach built as Claude Code skills that render locally with Python and ffmpeg, see the open source Claude video editor and the OPEN VIDEO EDIT update. If you want a shorter path from idea to finished short, the one line launch video skill for Claude Code is worth a read.

HyperEdit FAQ

Is HyperEdit free and open source?

Yes. The source is published on GitHub at github.com/kevinbadi/hyperedit as Kevin’s own open source project. You clone it, run it locally and keep your footage on your own machine.

How do I install and run HyperEdit locally?

The repository lists two commands: npm install and npm run dev. Run them in the cloned project folder and open the dev server in your browser. The repo also links a Discord community for setup help.

Does HyperEdit work with Claude or MCP video editor workflows?

Yes, as a local project a coding agent can drive. The agents are already part of the design and Remotion components are React code an agent can read and edit. For published output, Creator OS provides a hosted MCP server and a CLI so the same agent can schedule and post the finished clip.

What is Jev in HyperEdit?

Jev is the engine the project names as powering its agents. The repository describes HyperEdit as having Obsidian agents baked in and powered by Jev. Configure it as the project directs before you rely on agent planned edits.

Is HyperEdit better than other open source video editors for AI editing?

It is built for a different job. If you want a timeline you drag by hand, a traditional open source editor fits better. If you want the edit planned from notes and rendered from code, HyperEdit’s FFMPEG, Remotion and Obsidian stack is designed for exactly that.

Get Started With HyperEdit and Creator OS

Clone HyperEdit on GitHub, run npm install and npm run dev, and render one clip before you connect anything else. When the file looks right, the next step is getting it in front of people.

That is where Creator OS comes in. Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Your agent can create a post, attach the rendered video and schedule it, using tools like create_post and upload_media_from_url. With the CLI, upload the file first so it gets a media id, then schedule the post:

creatoros media:upload ./out/update.mp4

creatoros posts:create --text "Pricing update in 45 seconds" \
  --platforms instagram,tiktok,youtube \
  --media med_abc123 \
  --scheduledAt 2026-03-02T15:00:00Z --timezone America/Toronto

The media:upload command returns a med_ id. posts:create –media takes that id or a hosted URL, not a path on your disk, so upload before you schedule. Read the posting options at creatoros.ca/docs/post-options before you schedule, since each platform applies its own rules. And watch the walkthroughs on the KevBuildsApps YouTube channel, including the launch video, to see the local build and the agent workflow side by side.

Get started for $0
Free trial for new subscribers. Cancel anytime.