generated: '2026-07-19'
method: derived
source: openapi/kalpa-labs-openapi-original.json
entities:
- name: Conversation
  description: An ordered list of turns, oldest first; the last (open) turn is completed by the model.
  schema: ConverseRequest
- name: ConversationTurn
  description: One turn — speaker (positional role label), optional text, optional audio_wav_b64.
  schema: ConversationTurnModel
- name: ConverseReply
  description: The completed open turn — speaker, authored/rendered text, and audio.
  schema: ConverseReply
- name: AudioPayload
  description: Generated audio — format, sample_rate, audio_quality, base64 data_b64.
  schema: AudioPayload
- name: ModelCard
  description: A public model registry entry — id, display_name, modes, speakers, default.
  schema: ModelCard
- name: VoiceCard
  description: A stored voice usable with POST /v1/tts/{voice_id}.
  schema: VoiceCard
- name: Usage
  description: Per-response metering — input_chars, input_audio_seconds, output_audio_seconds.
  schema: Usage
- name: UsageSummary
  description: Running per-key totals returned by GET /v1/usage.
  schema: UsageSummaryResponse
relationships:
- from: Conversation
  to: ConversationTurn
  kind: has_many
  via: conversation[]
- from: ConverseReply
  to: AudioPayload
  kind: has_one
  via: audio
- from: ConverseReply
  to: Usage
  kind: has_one
  via: usage
- from: ConversationTurn
  to: AudioPayload
  kind: has_one
  via: audio_wav_b64 (input reference/spoken history)
- from: ConversationTurn
  to: ModelCard
  kind: belongs_to
  via: speaker labels are defined per ModelCard.speakers
notes: >-
  Voices (VoiceCard) and models (ModelCard) are the two reference registries;
  every generation request selects a model by id and renders speakers by the
  labels that model advertises.