trx: transcribe audio and video from the CLI

Agent-first CLI for audio and video transcription with local Whisper. Download, clean, and transcribe media from URLs or local files with machine-readable output.

trx downloads media with yt-dlp, cleans audio with ffmpeg, and transcribes it using a local Whisper model. It outputs .wav, .srt, .txt, and JSON, and it is designed for AI agents first: JSON output, dry-run validation, field filtering, and runtime schema introspection.

Install

bun add -g @crafter/trx
trx init

Quick start

Transcribe a local file:

trx recording.mp4

Transcribe from a URL (YouTube and anything yt-dlp supports):

trx "https://youtube.com/watch?v=..."

Agent-friendly usage

Machine-readable output, designed to be called from scripts and AI agents:

# JSON output
trx transcribe video.mp4 --output json

# Only the text field
trx transcribe video.mp4 --fields text --output json

# Validate without executing
trx transcribe video.mp4 --dry-run --output json

# Inspect the command schema at runtime
trx schema transcribe

Language support

trx transcribe video.mp4 --language es

Model sizes

Whisper models range from tiny (75MB) to large (3GB). Pick the size that fits your accuracy and speed needs during trx init.