Vida

Experimental: read the status MIT License Runs on your computer

The agent makes the draft. You make the decisions.

Vida is a small local video editor for episodes made from several recordings. An agent writes an editable project file. You open the same file in the Vida editor, look at each cut, and change what you do not like.

Install Vida View the code on GitHub
The Vida editor with a project named Demo episode. Left: four sources and three named selects. Center: the source view of intro take 2 with an audio waveform, and the sequence preview with the caption 'then it runs validate and fixes every error.' Right: the whisper.cpp transcript. Bottom: a sequence of four colored clips with waveforms, and the clip inspector with In and Out times.
Real screenshot of Vida. The agent wrote the sequence and the selects. The footage is generated demo footage with a synthetic voice. The transcript is from whisper.cpp.

How it works

One project file goes from the agent to you, in order.

  1. 1 · Agent Writes a draft It reads the transcripts and writes clips and named selects to Episode (agent draft).vida.json.
  2. 2 · Agent Validates it vida validate reads each source with ffprobe and names each bad clip. The agent fixes every error.
  3. 3 · You Review in the editor Open the draft in Vida. Change takes, trim, split, reorder, and correct caption words.
  4. 4 · You or the agent Export One MP4 with SRT and VTT captions. Caption times move with each cut.

Vida does not choose the edits. The agent proposes the cuts, and it does not always choose the best take. Vida checks that each cut is possible and renders it exactly. You decide what is final.

Review and adjust

The draft is not a finished video. It is a project that you can edit.

  • Each source keeps its transcript. Click a line or a word to go to that time.
  • Double-click a word to correct it. The word keeps its time, so the captions stay in sync.
  • Alt+click a filler word to remove it from the captions. It stays in the transcript.
  • The agent can propose ranges as named selects. You add the ones that you want with one click.
The Vida transcript panel for intro take 1. The word 'Bye' is open for correction and now reads 'Hi'. The filler word 'um' is struck through because it is off in the captions. Below: 'Wait, let me start again.' and a 'No speech' row.
whisper.cpp heard “Bye”. A person corrects it to “Hi” and turns off “um”.

Safe hand-off

The agent and you can work on the same project.

  • Vida checks the project file every 2 seconds. When the file changes, a bar shows Reload from disk.
  • After a reload, Undo brings back your earlier version.
  • Save sends the revision of your version. If the file on disk is newer, the save fails and the file stays the same.
  • This check is not a file lock. Agents write to a new draft file, not to a file that you have open.
The Vida editor with a yellow bar below the top bar: 'The project file changed on disk. Reload it to see the new version.' Buttons: Reload from disk, Save as, Keep editing. The Selects list shows 4 selects, and the transcript shows 'um' struck through.
An agent added a select while the project was open in Vida. The person decides when to reload.

Features

A small editor for one job: cut several recordings into one episode.

Several takes

Import screen recordings and camera files. Vida reads them in place. It does not copy, change, or delete them.

Non-destructive cuts

Trim to the frame, split at the playhead, drag to reorder, and remove clips. Undo and Redo.

Transcripts and captions

Cut by transcript lines. Correct each word. Silent parts show as “No speech” rows.

Named selects

Keep good ranges for later with a name. An agent can propose ranges this way.

Audio waveforms

See speech and silence in the source view and on each clip before you cut.

Keyboard editing

I O mark, A add, N select, S split, [ ] trim, J K L play.

Saved JSON projects

A .vida.json file keeps your edit decisions and transcripts. People and agents can read and write it.

Headless CLI

validate, captions, and export work without the browser and use the same timeline code.

MP4, SRT, VTT

H.264 and AAC video with captions. Audio stays in sync after many cuts.

For agents

An agent uses the same project format and the same rules as the editor.

  • Validation names each bad clip, for example a clip that ends after its video.
  • captions and export validate first. If the project is not valid, they write nothing.
  • Export never writes to the path of a source file.
  • Agents must not invent transcript text. They use whisper.cpp, the API provider, or SRT/VTT files.

Read the agent workflow and project format →

Output from the demo project
$ node bin/vida.js validate "Demo episode (agent draft).vida.json"
Valid: Demo episode
4 clips, 3 selects, 26.200 s, 1920x1080 at 30 fps, 6 caption cues
  source take1: intro-take-1.mp4, 12.000 s, 2 transcript lines
  source take2: intro-take-2.mp4, 13.000 s, 2 transcript lines
  source screen: screen-demo.mp4, 24.000 s, 3 transcript lines
  source outro: outro.mp4, 7.000 s, 1 transcript lines

$ node bin/vida.js export "Demo episode (agent draft).vida.json"
# writes exports/Demo episode (agent draft).mp4, .srt, and .vtt

Local by design

This website is only a homepage. The editor is not hosted. It runs on your computer.

What stays on your computer

local
  • The editor server listens on 127.0.0.1 only. Other computers cannot connect to it.
  • FFmpeg reads your files and renders the export on your computer.
  • There are no accounts, no cloud storage, and no publishing.
  • With local whisper.cpp, the audio stays on your computer.

What can leave your computer

optional
  • You can set up an OpenAI-compatible transcription API. When you click Transcribe (API), Vida extracts the audio of that video and uploads it to the provider that you configure.
  • Vida does not upload the picture. Vida itself makes no other network requests.
  • To upload nothing, use whisper.cpp or import SRT or VTT files that you already have.

While the server runs, any program on your computer can use it. It can list your folders, read video files, and write files where your user account can write. Run Vida only on a computer that you trust, and stop it when you do not use it.

Install

Vida is free and open source under the MIT License.

You need

  • Node.js 20 or newer.
  • FFmpeg with ffprobe. On macOS: brew install ffmpeg.
  • A Chromium-based browser. Other browsers are not tested.
  • Optional: whisper.cpp and a model, for local transcripts.
Terminal
git clone https://github.com/zainfathoni/vida.git
cd vida
npm start   # opens http://localhost:5178/

GitHub repository README and setup Transcription setup Rules for agents

Status: experimental

Vida was tested only on Linux, in Chromium, with generated test videos. It was not tested on macOS, Windows, or Safari, or with real camera and screen recordings. The API transcription option was tested only with a mock API. Keep backups of your project files.

Vida does not do multicam, transitions, audio level controls, automatic vertical crops, or motion graphics. It has no AI features of its own: the agent that you use makes the suggestions.