The ozen binary is built to ~/ozen/target/release/ozen. It works from any directory. Most commands do what a
panel button does, so you can script ozen or drive it from an agent.
| Command |
What it does |
ozen start |
Start recording (same as Start). |
ozen pause |
Pause; the transcriber stays loaded. |
ozen resume |
Resume after a pause. |
ozen stop |
Stop, finish transcribing queued audio (up to 2 minutes), then unload. |
ozen status |
Print recording, paused, stopping or stopped. |
ozen health |
Print one line per problem the panel warns about; nothing when all is well. See Troubleshooting. |
| Command |
What it does |
ozen show [N] |
Last N lines, with tagged speakers and fixed text. |
ozen look [N] |
Screenshot to ~/ozen/screen-small.png, then the last N lines (default 40). |
ozen fix <line-id> "right text" |
Correct a line’s text; empty restores it. Relearns hint words and corrections. |
ozen tag <line-id> "Name" |
Say who said a line; empty clears. Retrains. |
ozen ignore <line-id>... |
Mark lines as a voice to ignore. Retrains. |
ozen retrain |
Rebuild voiceprints, labels, ignored voices and accuracy from all tags. Tagging does this for you. |
| Command |
What it does |
ozen meetings |
Past meetings, newest first: id, start, minutes, lines, first words. |
ozen gather [--kev] ID... |
Write those meetings to ~/ozen/context/<now>/ and print the folder. --kev adds meetings Kev judges related. |
ozen live [--open claude|hermes] |
Write the current meeting to ~/ozen/context/live/, keep it updated every 15 s until the meeting ends, and print the folder. --open starts that agent there. |
ozen open DIR claude|hermes|finder |
Open a folder written by gather or live in that agent, or in Finder. |
ozen mcp |
Run the MCP server on stdio. |
| Command |
What it does |
ozen app |
Build and install ~/Applications/Ozen.app. |
ozen bar |
Build if needed, then open Ozen.app. |
| Command |
What it does |
ozen eval [--vocab 0,10,30] [--repeat 0,1,2] [--real] [--fresh] |
Score learning settings; see Tuning how fixes teach. |
uv run eval.py [N] |
Transcribe your last N real chunks with stock, Hebrew and Hebrew+vocabulary models, to compare on your own speech. Run from ~/ozen. |