What is editonny?
editonny, by Matheus Gomes, is an agent skill rather than an app. You ask Claude Code or Codex for an edit in plain language and the skill steers the agent through a fixed workflow: inspect the footage, propose every setting and wait for confirmation, cut pauses, breaths, mistakes and false starts against the waveform, add captions, adjust the audio, reframe to 9:16, 16:9, 1:1 or 4:5, and export a checked MP4 after you review a sample of 8 to 15 seconds.
The editing judgement comes from your agent on its existing plan; there is no separate LLM API. Transcription runs locally with mlx-whisper on Apple Silicon Macs and faster-whisper elsewhere, captions are checked with local OCR (Apple Vision or Tesseract), and rendering uses FFmpeg and HyperFrames 0.8.179 in a pinned headless browser. It bundles 145 motion scenes and 79 original clips. External image, voice or video generators are optional and need your explicit authorisation.
The repository was created on 19 October 2026 and has about 457 GitHub stars. First-party code is MIT-licensed; bundled models, fonts and media keep their own licences. There are no GitHub releases: the plugin manifest says 1.1.0, and the last commit was on 21 October 2026. The project homepage is the author's Instagram account.
Is editonny worth it?
editonny is worth watching but not yet worth relying on. Its design is sound: it treats an edit as a contract, asks you to confirm every setting before it cuts, keeps the original footage untouched, never overwrites an export, and blocks colour paths it does not support rather than guessing. For the job it targets, a creator turning a raw take into a captioned vertical clip, that discipline matters more than a long feature list.
It is narrower than its neighbours. ffmpeg-skill gives an agent 42 typed FFmpeg tools for almost any edit; editonny is about speech, pauses and captions, and adds a scene layer of 145 HyperFrames motion scenes that ffmpeg-skill lacks. OpenMontage aims at a whole production pipeline with paid generation at its centre, while editonny keeps everything local by default. Concat is a different kind of tool, a timeline editor with its own MCP server, for people who would rather drag clips than describe them.
The project's validation notes are unusually candid about maturity. The evidence covers SDR cut paths on macOS with Apple Silicon; Windows and Linux runtime behaviour, long-form throughput and arbitrary phones and cameras are listed as unvalidated, and native HDR overlays, Dolby metadata and multi-track audio are blocked. The issue tracker agrees, with a report that every transcription fails on Windows in 1.1.0 and another that a cut with no audio stream has no supported route.
The weight on one person shows. All 27 commits landed between 19 and 21 October 2026, with nothing since, and the notes record that code reviews were run by read-only Codex sessions. Setup also asks a lot: Python, uv, FFmpeg, Node.js 22.13 or later, a headless browser and a Whisper model, installed by pasting a long request into your agent, which then clones the repository and runs its installer. That is convenient, and it is also a habit worth questioning with any skill.
For a Mac user cutting short talking-head clips, it already handles the tedious part well; on any other platform, ffmpeg-skill's longer release record makes it the steadier choice today.
Where it earns its place
A solo educator on an Apple Silicon Mac who records ten-minute takes for short-form video
Turn a raw take into a 45-second captioned vertical clip
Call /editonny in Claude Code with the footage folder; it proposes settings, waits for a yes, cuts pauses and false starts, captions in one of four lettering styles and asks for approval of a short sample before the full export. Free beyond the existing agent plan.
A two-person marketing team at a B2B startup producing founder explainer videos
Add simple motion graphics where the founder lists steps or quotes numbers
Ask for scenes only where they explain the story; editonny picks from its bundled catalogue, such as stacked steps or a number spotlight, and renders them locally with HyperFrames. Sample figures and brands in scenes are placeholders and must be replaced with real facts before publishing.
A developer evaluating agent video tools for a Windows-based content team
Decide whether editonny can be rolled out now
Read docs/validation.md and the open Windows issues first: reported transcription failures in 1.1.0 and unvalidated Windows and Linux runtimes point to a Mac-only pilot, or to ffmpeg-skill for a mixed fleet.
Before you commit
- Setup
- Paste the install request from the README into Claude Code or Codex, or run python3 install.py --target both and then editonny.py doctor. Needs Python 3.9+ for the launcher (the engine uses 3.10 to 3.13 through uv), FFmpeg and ffprobe, Node.js 22.13+ or 24+, and Apple Vision or Tesseract for OCR.
- Integrations
- Claude Code (/editonny) and Codex ($editonny); HyperFrames 0.8.179 for rendering; mlx-whisper or faster-whisper for transcription. Exports can be finished in Shotcut or Kdenlive, but there is no timeline export.
- Data and privacy
- Transcription, OCR and rendering run locally, and the guide website never receives footage; what the agent reads goes to its model provider as usual. Work is kept under Downloads/editonny. MIT for first-party code; bundled models, fonts and media keep their own licences.
- Cost at scale
- No paid edition. The costs are your agent subscription and local compute; optional external generators bill separately and need your authorisation. One project render runs at a time.
- Limits
- Validated on macOS with Apple Silicon only; HDR overlays, Dolby metadata, multi-track audio and non-square pixels are blocked; Windows transcription failures are reported in 1.1.0; long-form throughput is unvalidated.
What’s good
- Confirms every setting before cutting; never overwrites originals or exports
- Local transcription with mlx-whisper or faster-whisper, no extra API key
- 145 bundled motion scenes for explanatory overlays
- Unusually candid validation notes
What’s not
- Twelve days old, one maintainer, no tagged releases
- Validated only on macOS with Apple Silicon; Windows failures reported
- Speech-led only: cuts without an audio stream have no supported route
- Heavy toolchain of FFmpeg, uv, Node.js, a headless browser and Whisper
Best for
Mac-based creators and small teams already on Claude Code or Codex who want speech-led clips cut and captioned without uploading footage.
editonny alternatives
editonny: common questions
- Is editonny worth it?
- A careful, local editing workflow for speech-led video that is twelve days old, tested on Macs and already stumbling on Windows. We rate it 3.4 out of 5.
- How much does editonny cost?
- Free and open source (MIT). Your Claude Code or Codex subscription is separate, and any optional external generator bills you directly.
- What are the downsides of editonny?
- Twelve days old, one maintainer, no tagged releases; validated only on macOS with Apple Silicon; Windows failures reported; speech-led only: cuts without an audio stream have no supported route; heavy toolchain of FFmpeg, uv, Node.js, a headless browser and Whisper.
- Who is editonny best for?
- Mac-based creators and small teams already on Claude Code or Codex who want speech-led clips cut and captioned without uploading footage.
- How much does editonny cost for a team or in production?
- No paid edition. The costs are your agent subscription and local compute; optional external generators bill separately and need your authorisation. One project render runs at a time.
- Is editonny safe for company data?
- Transcription, OCR and rendering run locally, and the guide website never receives footage; what the agent reads goes to its model provider as usual. Work is kept under Downloads/editonny. MIT for first-party code; bundled models, fonts and media keep their own licences.
- What are the best editonny alternatives?
- The closest alternatives we have reviewed are ffmpeg-skill, OpenMontage, Concat and HyperFrames.
Answers checked Oct 31, 2026.