Worked example: audio foundation

This is an illustrative example with placeholder candidates and scores, not a captured Hub search or runtime result. Actual output depends on current metadata.

This example shows the audio foundational modality (ASR / TTS / music). No audio-specific discovery roles are added; the profile seeds the existing general role plus ASR keywords.

CLI

python3 scripts/mlx-agent research \
  --domain "meeting transcription and summaries" \
  --modality audio --facet asr \
  --keyword whisper

Or rely on detection from domain text (transcription / whisper).

Resulting intent (shape)

{
  "domain": "meeting transcription and summaries",
  "roles": ["general"],
  "keywords": ["whisper", "asr", "transcription", "speech-to-text", "audio", "speech"],
  "modalities": ["audio"],
  "facets": ["asr"]
}

Pack excerpt

## Runtime preference

- Preferred: `lmstudio` (or `mlx_lm` when LM Studio is down)
- Alternates: `ollama` when detected
- Rationale: Audio packs prefer native MLX; Ollama is limited to curated tags.

## Modality foundations

- `audio` — Audio (ASR / TTS / music)
  - Facets: `asr` (ASR / speech-to-text)

## Candidates
...

Music or TTS packs use the same foundation with --facet music or --facet tts.