Skip to content

CLI reference

Authoritative help snapshots (generated): cli-help.md. When this page and the binary disagree, trust aurum --help / the snapshot.

aurum <AUDIO_FILE> [OPTIONS]
aurum models
aurum models recommend --profile balance
aurum transcribe <AUDIO_FILE> [OPTIONS]
aurum batch <INPUT> --output-dir <DIR> [OPTIONS]
aurum cleanup [TEXT_FILE] [OPTIONS]     # alias: aurum flow
aurum tts "Hello" --output-file out.wav
aurum tts models | voices | adapters | inspect | verify | add
aurum converse <AUDIO_FILE> --reply-text "Hello" -O out.wav
aurum doctor [--json]
aurum cache status | verify | repair
aurum support-bundle
aurum completions zsh
aurum man
aurum --help
aurum --version

Transcribe options

Flag Default Description
--provider local\|openrouter\|openai\|xai local ASR backend (grok config alias → xai)
--model <NAME> base (local) Local ggml name or reviewed remote model id (wins over --profile)
--profile speed\|balance\|quality off Intent mapping (JOE-1723); does not change the default when omitted
--language <CODE> auto Language hint or auto-detect
-o, --output txt\|srt\|json txt Output format
--output-file <PATH> stdout Write to file
--timestamps off Request segments (implied by srt)
--allow-unreliable-timestamps off Force SRT when the route does not guarantee timestamps
--openrouter-stt-mode auto\|chat\|transcriptions auto OpenRouter route only; unknown models fail closed in auto
--cleanup <style> config / raw raw | clean | bullets | professional | summary
--cleanup-provider <p> rules rules | openrouter
--cleanup-model <id> OpenRouter default LLM cleanup model
--cleanup-segments <p> auto auto | keep | clear | per-segment
-v, --verbose off Diagnostics on stderr

Batch options (aurum batch)

Flag Description
INPUT File or directory of audio
-O, --output-dir Transcripts + aurum-batch-manifest.json
--recursive Walk subdirectories
--resume / --retry-failed Resume / retry failed items
--dry-run Manifest only
--profile / --model Same semantics as transcribe
--json Machine-readable summary

Support bundle

aurum support-bundle -O aurum-support.json
aurum support-bundle --stdout

Redacted offline diagnostics only (no audio, transcripts, or API keys).

TTS options (aurum tts)

Flag Default Description
TEXT / - / --input-file — Exactly one UTF-8 text source
--provider local local | openrouter | openai | elevenlabs | xai
--model local kitten-nano-int8; remote uses provider default Local pack id or reviewed remote model id
--voice local Luna; remote uses provider default Local alias or remote voice id (no cross-provider remap)
--language en Language hint
-o, --output wav Only wav
-O, --output-file required Destination WAV path
--force off Overwrite existing non-empty file
--speaking-rate 1.0 Clamped 0.5..=2.0
--cleanup off Rules-only raw | clean
--timeout config / 120000 Milliseconds
--local-only off No download if pack missing (local only)
--emit-json off Honesty JSON on stdout
-v off Diagnostics

When only --provider is set for remote TTS, model/voice default to that provider’s reviewed defaults (not local Kitten/Luna). See TTS guide.

Cleanup-only options

Flag Default Description
TEXT_FILE stdin Input text
-s, --style clean* Cleanup style (* defaults to clean if config is raw)
--provider rules Backend
--model config OpenRouter model
-o txt\|json txt Output
--output-file stdout Write path

Exit codes

Code Meaning
0 Success
2 User error (missing file, bad model, missing API key)
3 Environment error (ffmpeg missing, I/O)
4 Provider error (download, network, inference)
1 Internal

Examples

aurum models
aurum interview.mp3 --model small-q5_1 --language en -o json
aurum interview.mp3 --cleanup clean --cleanup-segments keep
aurum interview.mp3 --provider openrouter --model openai/whisper-large-v3
aurum interview.mp3 --provider openai --model whisper-1
echo "um hello" | aurum cleanup -s clean
aurum cleanup notes.txt --style bullets -o json
aurum tts "Hello from aurum" --output-file /tmp/a.wav --emit-json
aurum tts --input-file prompt.txt -O /tmp/a.wav --voice Luna
aurum tts "Hello" --provider openai --model tts-1 --voice alloy -O /tmp/oai.wav
aurum tts models
aurum tts voices