Solocut
日本語 GitHub Download

Agentic Development Environment · Free & open source

The ADE for
video editing.

Shoot alone; your agent edits by template. Made for tech tutorials and talking-head videos, Solocut records your screen, camera and mic at once, then Claude Code, Codex or any terminal agent cuts, captions and lays out the video by your template. A decision model checks every cut frame by frame, and you give feedback by talking over the preview.

Get notified on GitHub Read the design

macOS · Windows · Linux · MIT license · YouTube / TikTok / Shorts / Reels

Screen, two cameras and mic, each on its own track.

An animated model of the app.

Works withClaude CodeCodexGemini CLIOpenCodeCursorGitHub CopilotYour own CLI

How it works

From one take to a finished video, without a timeline.

Every step is a command your agent runs. The template decides the order, and solocut next always tells the agent what to do now.

  1. Pick a template

    TikTok talk, YouTube tutorial, Shorts from a long video… Format, layout, cut policy and caption style are set.

  2. Record

    Screen, camera and mic at once, each on its own track. Add footage from another camera later.

  3. Ask your agent

    “Edit this.” Claude Code or Codex follows the template: sync, transcribe, rough cut, captions, layout.

  4. Decision model decides

    Cut or keep, which retake, where captions break. Only uncertain items come to you.

  5. Check every frame

    Cut boundaries, black frames, caption safe areas, faces, personal info on screen, mosaic tracking.

  6. Talk over the preview

    Circle what bothers you and say it. It reaches your agent as timecoded instructions. Then export.

Record

Screen and camera at once. Another camera, synced later.

Each source is saved on its own track, so the template can switch between picture-in-picture, split screen, camera only or screen only at any point.

  • Screen, cameras, mics and system audio recorded together on one clock
  • Import footage from a mirrorless camera or iPhone; it syncs to the frame by audio, with drift correction for long takes
  • Teleprompter, countdown, and hotkeys to mark “bad take” or “important” while you talk
  • Crash-safe: files are written every second
screen
camera
camera B
mic

✓ camera B synced · offset −128 frames · confidence 0.97

Templates

Decide once how your videos are made.

A template fixes the look, the editing rules written as prompts, and the pipeline. Any agent follows the same steps and the same quality bar, because Solocut generates the agent’s instructions from it.

  • Built-in: tech tutorial, tech short, talk for YouTube, talk for TikTok, Shorts from a long video
  • Tech templates zoom to where you click, show the keys you press and highlight the cursor
  • Shooting with several cameras? Write how to switch angles in plain words; your agent follows the prompt
  • solocut next enforces the steps and their pass conditions
# templates/tiktok-talk/template.yaml
id: tiktok-talk
canvas: { width: 1080, height: 1920, fps: 30 }
layouts: [camera-full, screen-top-camera-bottom]
cut:
  silence: { minMs: 300, padMs: 80 }
  fillers: [um, uh, like]
  retakes: keep-last
captions: { style: tiktok-bold, highlight: word, safeArea: tiktok }
audio: { loudness: -14, truePeak: -1 }
checks: [cut-boundaries, caption-safe-area, pii-on-screen]
pipeline:
  [ingest, sync, transcribe, rough-cut, captions,
   layout, preview, check, review, final]

Terminal

Your agent, your subscription, your terminal.

Solocut runs agent CLIs in real terminals inside the app. Switch between them in tabs; the project state lives in files, so any agent can pick up where another stopped.

  • Any CLI agent

    Claude Code, Codex, Gemini CLI, OpenCode, Cursor, GitHub Copilot, or your own command.

  • Switch and hand off

    Hit a usage limit? The next agent opens with the handoff note and continues.

  • Always up to date

    Solocut updates itself, signed, from Cloudflare. It also tells you when your agent CLIs have a new version.

  • Headless CLI

    solocut works without the app, too: in your own terminal or in CI.

  • Every edit is undoable

    Edits are operations in a log. See who changed what and why, and roll back.

  • Originals stay untouched

    Recorded footage is read-only. Everything else can be rebuilt.

Decision model

A final check on every frame that matters.

Machine checks run first (black frames, freezes, loudness, specs). Then a decision model looks at the frames around every cut, every caption change and every mosaic, and answers typed questions with a probability. Confident answers are applied; uncertain ones come to you.

cut 00:12:04eyes open · pass 0.94
caption c18in safe area
cut 00:41:07mid-word? 0.58 → you
frame 01:02:11email visible → mosaic
  • Runs locally with Ollama, or with your own API key (TypeSafe, OpenAI, Anthropic, Gemini, Workers AI)
  • Spending limits per project and per day, with a log of every call
  • Only frames and transcript text are sent, and only after you agree

Mosaic

Personal info, hidden and tracked.

Emails, notifications, API keys, phone numbers, bystanders’ faces and license plates are detected and tracked, then pixelated, blurred or filled. Masks follow the source, so they stay in place when you change the layout or reframe to 9:16.

// timeline.json
"masks": [{
  "kind": "pixelate",
  "reason": "pii:email",
  "source": "t1.screen",
  "track": [
    { "frame": 120, "box": [412, 88, 260, 28] },
    { "frame": 180, "box": [412, 40, 260, 28] }
  ],
  "confidence": 0.93
}]

Feedback by video

Don’t type the fix. Point at it.

Play the preview, talk, pause and circle. Solocut transcribes on your machine and turns each remark into a timecode, a frame with your pen, your words and the exact caption or clip you meant. One click sends it to your agent.

Benchmark

Paste a URL. Borrow the pattern, not the video.

Open a reference video — your own, or one you have permission to study — in the built-in browser. While you play it, Solocut measures how captions appear, where sound effects land, how loud the music sits under the voice and how fast the cuts come, then turns it into a template you can use. Nothing is downloaded or saved; only numbers are kept.

youtube.com/watch?v=…

Captions
bottom 22% · 14 chars · word highlight · yellow emphasis
Cuts
2.4 s average · 14 jump cuts/min · punch-in ×1.12
Music
lo-fi · 88 BPM · −21 dB under voice · ducking
Sound effects
9/min · pop on emphasis · whoosh on new section
Voice
380 chars/min · −15 LUFS

Apply to talk-tiktok

Libraries on demand

The best open tools, downloaded when you need them.

Solocut stays small. FFmpeg, whisper.cpp and the rest are fetched on first use, pinned to a version and verified by hash. Tools with special license terms ask for your consent first.

FFmpegcut, compose, captions, loudness
whisper.cppon-device transcription
RemotionReact motion graphics
license terms apply
HyperFramesHTML motion graphics
RevideoTypeScript motion graphics
MediaPipefaces, reframing, mosaic
PySceneDetectscene detection
OpenTimelineIOto Premiere / Resolve / FCP

Why Solocut

Built for one person.

  • Shoot 20 minutes. One prompt.

    Say “finish the video” and your agent runs every step of the template, up to a publish-ready video with thumbnails and a description.

  • Ready-made prompts

    Rough cut, fix feedback, make 3 Shorts, write the post… One click each, and every prompt is yours to edit in JSON.

  • Publish and announce

    Upload to YouTube and TikTok from the editor, then post “new video” to X, Instagram, Threads and Facebook. Nothing goes out until you approve.

  • Thumbnails with image generation

    Pick a frame and generate thumbnails with your Codex or Gemini subscription, or an OpenAI or Google API key. Titles are typeset so Japanese text never breaks.

  • Sound effects and music, on tap

    A library of sound effects, BGM and illustrations you are free to use. Each one downloads the moment you or your agent picks it, with credits written for you.

  • Your past videos, at hand

    Every finished video and its timeline stay searchable. Ask “make it like the one from last month” and your agent follows that edit.

  • Every setting is JSON

    Settings live in a JSON file with a public schema. Edit it by hand or just ask your agent to change it.

  • Free and open source

    MIT license. No account, no watermark, no subscription.

  • Private by default

    Video and audio never leave your machine unless you choose a cloud model for checks.

  • macOS, Windows, Linux

    The same app and the same results on all three.

  • YouTube and TikTok ready

    Export presets for YouTube, TikTok, Shorts and Reels with safe areas, loudness and metadata.

  • Japanese-grade captions

    Natural line breaks with BudouX, word highlight, emphasis and term correction.

  • What you see is what you get

    Preview and final export come from the same render engine.

FAQ

Is it really free?

Yes. Solocut is MIT-licensed. The editing intelligence comes from the agent subscription you already have (Claude Code, Codex…), and the decision model can run locally.

Can I use it today?

Not yet. Solocut is being designed and built in the open. Watch the GitHub repository to get the first release.

Do I need to know how to edit?

No. You record and talk. The template and your agent do the editing; you review the result.

Is Remotion required?

No. The render engine is FFmpeg. Remotion is an optional add-on for motion graphics, and Solocut shows its license terms before installing it. HyperFrames and Revideo are alternatives.

Which platforms can I export for?

YouTube (16:9, up to 4K), TikTok, YouTube Shorts and Instagram Reels (9:16), with captions, chapters and metadata.

Can it post for me?

Yes. Solocut uploads to YouTube and TikTok and announces the video on X, Instagram, Threads and Facebook, always after you check and approve the post.

Shoot alone. Ship every day.

Solocut is in development. Follow along on GitHub.

Get notified on GitHub Download