Cuthol is out — local video editor built for AI agents

Hey Product Hunt,

I’m Kutay, maker of Cuthol.

Cuthol is an open-source local NLE that AI agents can drive through MCP or CLI. Your agent cuts, captions, mixes and renders video on your machine, and checks the edit by reading text (transcripts, timings, geometry) instead of guessing from pixels.

Why it exists

Most people using agents for video end up regenerating the whole composition every time they want a small change. That burns tokens and limits fast. Cuthol keeps a persistent timeline. You edit the timeline, you don’t regenerate the video.

What you can do with it

• Cut pauses and filler words with word-level timing

• Generate captions that stay in sync

• Add motion graphics (HyperFrames) on lanes

• Mix audio, normalize loudness, duck under speech

• Verify the edit before render (the agent can look at real frames and catch issues itself)

Early numbers from real sessions

After setting up extensions (reusable assets/models), subsequent videos got 4.6×–12× cheaper per minute. A late change (cutting one sentence + two visual fixes) on a 6:35 edit cost ~$4.67 and took ~2 minutes — no composition was rewritten, timeline just updated.

It’s still very early. Local, open-source, runs on your machine. Node ≥ 22.5 + ffmpeg 8.x + Chrome (for HyperFrames).

Repo: https://github.com/HypeAITR/Cuthol

If you already use agents for video (Claude Code, Cursor, Codex, etc.) and are tired of regenerating everything, give it a try and tell me what breaks. Especially interested in feedback on the extension system and speech cleanup.

Thanks.