engines.yaml

# ORAM Engine Catalog
# declarative engine registry for future YAML-based loading
# format: list of engine specs that can be auto-registered
#
# this file is NOT loaded in Sprint 1 — it's a reference for
# the planned YAML loader in the EngineRegistry.

engines:
  # ── ElevenLabs (cloud, premium) ──
  - id: elevenlabs-sfx
    provider: elevenlabs
    label: ElevenLabs Sound Effects
    mode: cloud
    capabilities:
      - text_to_sound_effect
    requiresApiKey: true
    maxDurationSeconds: 22
    costPerSecond: 40
    latencyProfile: medium

  - id: elevenlabs-tts
    provider: elevenlabs
    label: ElevenLabs Voice
    mode: cloud
    capabilities:
      - text_to_speech
      - speech_to_speech
    requiresApiKey: true
    supportsStreaming: true
    supportsAudioInput: true
    maxDurationSeconds: 300
    costPerSecond: 30
    latencyProfile: fast

  - id: elevenlabs-music
    provider: elevenlabs
    label: ElevenLabs Music
    mode: cloud
    capabilities:
      - text_to_music
    requiresApiKey: true
    supportsStreaming: true
    maxDurationSeconds: 600
    costPerSecond: 50
    latencyProfile: slow

  - id: elevenlabs-scribe
    provider: elevenlabs
    label: ElevenLabs Scribe
    mode: cloud
    capabilities:
      - speech_to_text
      - audio_analysis
    requiresApiKey: true
    supportsAudioInput: true
    maxDurationSeconds: 3600
    costPerSecond: 10
    latencyProfile: medium

  # ── Stable Audio (cloud via fal) ──
  - id: stability-stable-audio-3
    provider: stability
    label: Stable Audio 3 Large
    mode: cloud
    capabilities:
      - text_to_music
      - text_to_sound_effect
      - audio_to_audio
      - audio_inpainting
      - audio_continuation
      - lora_mixing
      - audio_latent
    requiresApiKey: true
    supportsAudioInput: true
    supportsSeed: true
    maxDurationSeconds: 380
    costPerSecond: 0
    latencyProfile: slow

  - id: stable-audio-25-fal
    provider: fal
    label: Stable Audio 2.5 via fal
    mode: cloud
    capabilities:
      - text_to_music
      - text_to_sound_effect
      - audio_to_audio
    requiresApiKey: true
    supportsAudioInput: true
    supportsSeed: true
    maxDurationSeconds: 47
    costPerSecond: 5
    latencyProfile: medium

  # ── Local Runners (no API key needed) ──
  - id: stable-audio-3-local
    provider: local
    label: Stable Audio 3 Local Service
    mode: local
    capabilities:
      - text_to_music
      - text_to_sound_effect
      - audio_to_audio
      - audio_inpainting
      - audio_continuation
      - lora_mixing
      - audio_latent
    requiresApiKey: false
    supportsAudioInput: true
    supportsSeed: true
    maxDurationSeconds: 380
    costPerSecond: 0
    latencyProfile: slow

  - id: tangoflux-local
    provider: local
    label: TangoFlux Local
    mode: local
    capabilities:
      - text_to_sound_effect
    requiresApiKey: false
    maxDurationSeconds: 30
    costPerSecond: 0
    latencyProfile: slow

  - id: kokoro-local
    provider: local
    label: Kokoro Local TTS
    mode: local
    capabilities:
      - text_to_speech
    requiresApiKey: false
    maxDurationSeconds: 60
    costPerSecond: 0
    latencyProfile: medium

  - id: whisper-local
    provider: local
    label: Whisper.cpp Local
    mode: local
    capabilities:
      - speech_to_text
    requiresApiKey: false
    maxDurationSeconds: 3600
    costPerSecond: 0
    latencyProfile: slow

  - id: essentia-local
    provider: local
    label: Essentia Audio Analysis
    mode: local
    capabilities:
      - audio_analysis
    requiresApiKey: false
    maxDurationSeconds: 3600
    costPerSecond: 0
    latencyProfile: fast