Free · no install · no watermark

A full video editor, right in your browser.

Open a tab and start cutting: timeline, keyframes, titles, captions, 4K export — free, no account, no watermark. Everything runs on your machine, so your footage never uploads anywhere. And when you'd rather not do the busywork, your AI agent can drive the same editor.

my-cut.flyExport
cut to the beat

scale

opacity

speed

00:12.4 / 00:48.0
v1
a1
cc

What's in the box

30 features. 24 of them free.

All of it opens right in the browser — no trial clock, no watermark. Only the six marked Pro are paid.

See the full comparison

See it work

Two chores PiCut does for you.

One click in the editor — or one command from your AI agent. Same recipe, same result.

Made with PiCut
before
after

Cut the silences

Every pause, cough and dead stretch found and cut, gaps closed — the whole talking-head cleanup in one pass, and one undo if you change your mind. (A pure RMS analysis driving one undoable op batch.)

$ picut cut-silences
Made with PiCut
we run the render on your machine
cc

Auto-captions, burned in

Your words appear on screen, timed and burned into the export — transcribed on your machine, or from your own SRT. Unlimited, no per-minute meter. (Whisper speech-to-text feeding a caption track.)

$ picut analyze --transcribe && picut captions build

Illustrated above. Screencasts rendered with PiCut land at launch.

For AI agents

The same editor, in your agent's hands.

PiCut ships a CLI and an MCP server, so Claude Code — or any MCP agent — edits the very project you see: 21 typed ops, each one validated, applied atomically, and undoable. There's no LLM inside PiCut; your agent brings the brain.

Install in 60 seconds

  1. perceive

    PerceivePiCut analyzes media into time-indexed facts — silences, scenes, transcripts — read back as plain JSON your agent can reason over.

  2. act

    ActYour agent emits a batch of typed ops. PiCut validates every one, applies them atomically, and rejects bad input with an error the agent can retry.

  3. prove

    ProveA deterministic engine renders the result to a real MP4 on your machine — WebCodecs via headless Chrome, bit-identical to the web preview.

picut · mcp
agent → apply_ops
[{ "op": "splitClip",  "clipId": "c1", "atSec": 12.4 },
 { "op": "removeClip", "clipId": "c1b" },
 { "op": "moveClip",   "clipId": "c2", "startSec": 12.4 }]
picut ✓3 ops applied · silence removed · timeline tightened
agent → render -o out.mp4
picut ✓out.mp4 · 1080p · 00:48

Why PiCut

Built for agents, honest about compute.

The agent is the brain — not us

No LLM, no API key, no planner inside PiCut. Your agent brings the intelligence; PiCut is the typed hand it edits with. The op schema plus a project summary is the whole contract.

Your machine, your media

All decode, render, and analysis run locally through WebCodecs. Nothing leaves your machine except the audio slice you explicitly send to hosted captions.

One flat price, unlimited render

Because compute is local, there are no per-minute credits to meter. The CLI and web editor are free — one subscription unlocks hosted AI (captions + translation) and smart edit across both.

Every edit is undoable

Edits are typed ops appended to a log; undo is deterministic replay. It doubles as an audit trail of exactly what the AI did.

One flat plan. Unlimited local render.

Rivals meter render minutes and transcription. PiCut runs that compute on your machine, so it stays flat and unlimited. The free tier is fully usable — the op surface is never gated.

$19/mo· free tier, no login wall

See pricing