CLI help snapshot¶

Aurum is private speech I/O on your machine:
• STT — audio → text (whisper.cpp local by default; opt-in openrouter|openai|xai)
• cleanup — post-transcript flow styles (rules or OpenRouter)
• TTS — text → mono WAV (local ONNX by default; opt-in openrouter|openai|elevenlabs|xai)
• batch — resumable multi-file transcription
• converse — one file-in / WAV-out conversation turn (no mic)

Quick start:
 aurum meeting.m4a
 aurum meeting.m4a --cleanup clean
 aurum tts "Hello from aurum" --output-file /tmp/a.wav
 aurum converse talk.wav --reply-text "Hello" -O /tmp/out.wav
 aurum cleanup --style bullets < notes.txt
 aurum batch ./lectures -O ./out
 aurum models
 aurum models recommend --profile balance
 aurum tts voices
 aurum support-bundle
 aurum --help

Usage: aurum [OPTIONS] [AUDIO_FILE]
       aurum <COMMAND>

Commands:
  models          List local whisper models and cache status [alias: list-models]
  transcribe      Transcribe an audio file (default when AUDIO_FILE is given positionally) [alias: t]
  cleanup         Clean existing text (stdin or file) without re-transcribing [alias: flow]
  tts             Synthesize speech from text (local ONNX TTS → mono WAV)
  batch           Bounded resumable multi-file transcription (JOE-1726)
  converse        One file-in / WAV-out conversation turn (no microphone). Boxed: Windows default stack is 1 MiB; this variant is large (clap + live)
  cache           Inspect and verify local model/voice-pack cache (JOE-1592)
  doctor          Read-only system, config, cache, and capability diagnostics (JOE-1628)
  support-bundle  Privacy-safe support bundle for issue reports (JOE-1728)
  completions     Generate shell completions (bash/zsh/fish/powershell/elvish)
  man             Print a man page to stdout (JOE-1720)
  help            Print this message or the help of the given subcommand(s)

Arguments:
  [AUDIO_FILE]
          Audio file to transcribe

Options:
      --provider <PROVIDER>
          Transcription provider (registry id; default `local`)

      --model <NAME>
          Local model name (tiny-q5_1, base, …) or reviewed remote model id. Overrides --profile when both are set

      --profile <PROFILE>
          Intent profile: speed | balance | quality (JOE-1723). Ignored if --model is set

      --language <CODE>
          Language code (e.g. en, fr) or "auto"

  -o, --output <txt|srt|json>
          Output format

          [possible values: txt, srt, json]

      --output-file <PATH>
          Write output to this path instead of stdout

      --timestamps
          Include timestamps where available (implied by srt)

      --allow-unreliable-timestamps
          Allow SRT/timestamp output when the route does not guarantee reliable timings

      --openrouter-stt-mode <auto|chat|transcriptions>
          OpenRouter STT path: auto | chat | transcriptions (JOE-1586)

      --cleanup <raw|clean|bullets|professional|summary>
          Post-transcript cleanup style (default: raw = off)

      --cleanup-provider <rules|openrouter>
          Cleanup backend: on-device rules (default) or OpenRouter LLM

      --cleanup-model <MODEL>
          Model id when using --cleanup-provider openrouter

      --cleanup-segments <auto|keep|clear|per-segment>
          Segment policy after cleanup: auto | keep | clear | per-segment

  -v, --verbose
          Verbose diagnostics

  -h, --help
          Print help (see a summary with '-h')

  -V, --version
          Print version

List local whisper models and cache status

Usage: aurum models [COMMAND]

Commands:
  recommend  Explain an intent profile and the model it resolves to (JOE-1723)
  help       Print this message or the help of the given subcommand(s)

Options:
  -h, --help  Print help

Bounded resumable multi-file transcription (JOE-1726)

Usage: aurum batch [OPTIONS] --output-dir <DIR> <INPUT>

Arguments:
  <INPUT>  Audio file or directory of audio files

Options:
  -O, --output-dir <DIR>
          Directory for transcripts and the batch manifest
      --recursive
          Recurse into subdirectories when INPUT is a directory
      --resume
          Exact-match resume only (full source/output SHA-256 + operation fingerprint)
      --retry-failed
          Retry items marked failed or interrupted when resuming
      --reprocess-changed
          Opt in to reprocessing stale source/config/output items
      --verify-only
          Report resume decisions without transcription
      --dry-run
          Dry-run: write/update the manifest only (no transcription)
      --provider <PROVIDER>
          Transcription provider (validated against the provider registry)
      --model <NAME>
          Explicit model id (overrides --profile)
      --profile <PROFILE>
          Intent profile: speed | balance | quality (ignored when --model is set)
      --language <CODE>
          Language code or auto
  -o, --output <txt|srt|json>
          Output format for each item [possible values: txt, srt, json]
      --timestamps
          Include timestamps (implied by srt)
      --allow-unreliable-timestamps

      --openrouter-stt-mode <auto|chat|transcriptions>

      --cleanup <raw|clean|bullets|professional|summary>

      --cleanup-provider <rules|openrouter>

      --cleanup-model <MODEL>

      --cleanup-segments <auto|keep|clear|per-segment>

      --json
          Emit final summary as JSON on stdout
  -v, --verbose

  -h, --help
          Print help

One file-in / WAV-out conversation turn (no microphone). Boxed: Windows default stack is 1 MiB; this variant is large (clap + live)

Usage: aurum converse [OPTIONS] [AUDIO_FILE]

Arguments:
  [AUDIO_FILE]  Audio file for a single user turn (omit when using `--mic`)

Options:
      --mic                      Use the default microphone and speakers (half-duplex CLI loop)
      --provider <PROVIDER>      STT provider (registry id; default `local`)
      --model <NAME>             STT model id (local ggml name or reviewed remote id)
      --language <CODE>          STT language (default from config / auto)
      --tts-provider <PROVIDER>  TTS provider (registry id; default `local`)
      --tts-model <NAME>         TTS model id
      --voice <NAME>             TTS voice id (required for ElevenLabs; never remapped from Luna)
      --reply-text <TEXT>        Agent reply text (mutually exclusive with `--reply-file` and `--llm-provider`)
      --reply-file <PATH>        Read agent reply UTF-8 from this file
      --llm-provider <PROVIDER>  Chat backend for the agent turn: `openai` | `openrouter` | `xai`. Never selected just because a key is set
      --llm-model <NAME>         Chat model id (defaults: openai=`gpt-4o-mini`, openrouter=`google/gemini-2.5-flash-lite`; xAI requires this flag)
      --llm-system <TEXT>        Optional system prompt override for `--llm-provider`
  -O, --output-file <PATH>       Write agent WAV here (required for file mode; optional with `--mic`)
      --force                    Overwrite an existing non-empty output file
      --local-only               Reject remote STT/TTS before encode/upload
      --emit-json                Honesty JSON on stdout (no PCM). Incompatible with `--stdio`
      --stdio                    JSONL sidecar: host owns mic/brain (`transcribe`/`synthesize` paths). Does not require `--mic` or a file
      --endpoint-silence <SECS>  Silence that ends a `--mic` turn after a full utterance (seconds, default 0.7)
      --thinking-pause <SECS>    Extra `--mic` wait while the utterance is still short (seconds, default 1.2). Covers a ~1 s mid-thought pause. Still RMS energy, not a neural VAD
  -v, --verbose                  Verbose diagnostics
  -h, --help                     Print help

Read-only system, config, cache, and capability diagnostics (JOE-1628)

Usage: aurum doctor [OPTIONS]

Options:
      --json  Emit JSON report
  -h, --help  Print help

Inspect and verify local model/voice-pack cache (JOE-1592)

Usage: aurum cache [OPTIONS] <COMMAND>

Commands:
  status  Cheap size/existence inventory (no full hashing)
  verify  Full digest verification for STT pins; quarantines bad artifacts (no network)
  repair  Re-download a model through the verified path (requires network unless --local-only)
  help    Print this message or the help of the given subcommand(s)

Options:
      --json  Emit JSON instead of a table
  -h, --help  Print help

Synthesize speech from text (local ONNX TTS → mono WAV)

Usage: aurum tts [OPTIONS] [TEXT]
       aurum tts <COMMAND>

Commands:
  models    List local TTS models and cache status
  voices    List local TTS voices and cache status
  adapters  List supported TTS adapters (not bare ONNX loaders)
  inspect   Read-only inspect of a model pack or manifest (no network)
  verify    Full digest + adapter + conformance preflight for a pack
  add       Propose adding a local pack (dry-run by default; never auto-trusts HF)
  help      Print this message or the help of the given subcommand(s)

Arguments:
  [TEXT]  Text to speak. Use `-` to read UTF-8 from stdin

Options:
      --input-file <PATH>     Read UTF-8 text from this file instead of positional TEXT
      --provider <PROVIDER>   TTS provider (registry id; default `local`)
      --model <NAME>          TTS model id (default from config / kitten-nano-int8)
      --voice <NAME>          Voice id (default from config / Luna)
      --language <CODE>       Language code (default: en)
  -o, --output <wav>          Output container (WAV only) [possible values: wav]
  -O, --output-file <PATH>    Write WAV to this path (required for synthesis)
      --force                 Overwrite an existing non-empty output file
      --speaking-rate <RATE>  Speaking rate multiplier (clamped 0.5..=2.0)
      --cleanup <raw|clean>   Optional rules-only cleanup style before synth (`clean` min)
      --timeout <MS>          Wall-clock timeout in milliseconds
      --emit-json             Print honesty JSON metadata on stdout (audio only in file)
      --local-only            Fail if the voice pack is not already cached (no download)
      --pack-dir <DIR>        Local model-pack directory (manifest + artifacts). Not a bare .onnx path
      --allow-unverified      Allow trust=local_unverified for `--pack-dir` (unsupported for production)
  -v, --verbose               Verbose diagnostics
  -h, --help                  Print help

Clean existing text (stdin or file) without re-transcribing

Usage: aurum cleanup [OPTIONS] [TEXT_FILE]

Arguments:
  [TEXT_FILE]  Optional text file (default: read stdin)

Options:
  -s, --style <raw|clean|bullets|professional|summary>
          Cleanup style (default: clean for this subcommand, or config)
      --provider <rules|openrouter>
          Cleanup backend: rules (default) or openrouter
      --model <MODEL>
          Model when provider is openrouter
  -o, --output <txt|json>
          Output format for structured result [possible values: txt, json]
      --output-file <PATH>

  -v, --verbose

  -h, --help
          Print help

Generate shell completions (bash/zsh/fish/powershell/elvish)

Usage: aurum completions <SHELL>

Arguments:
  <SHELL>  Shell: bash | zsh | fish | powershell | elvish

Options:
  -h, --help  Print help

Print a man page to stdout (JOE-1720)

Usage: aurum man

Options:
  -h, --help  Print help

Privacy-safe support bundle for issue reports (JOE-1728)

Usage: aurum support-bundle [OPTIONS]

Options:
  -O, --output-file <PATH>  Output path (default: aurum-support-<unix>.json in cwd)
      --notes <TEXT>        Optional untrusted notes (do not paste secrets, transcripts, or private paths). Notes are scrubbed for common token/path shapes but are not privacy-guaranteed
      --stdout              Print JSON to stdout instead of writing a file
  -h, --help                Print help