generated: '2026-09-06'
method: derived
source: openapi/actionpower-daglo-cloud-api-openapi.yml, grpc/actionpower-speech.proto
note: >-
  The contract defines no components.schemas at all — every request and response schema is inlined at the
  operation. The graph below is derived from those inline shapes and from the rid handle that ties the
  asynchronous operations together. There is no persistent, addressable resource model here: the only
  server-side entity a caller can name is a job, identified by its rid, and there is no list, update or
  delete operation for one.
entities:
  - name: TranscriptionJob
    id_field: rid
    id_format: UUID string
    created_by: [post-stt-v1-async-transcripts]
    read_by: [get-stt-v1-async-transctips-rid, get-stt-v1-async-transctips-rid-format]
    lifecycle_field: status
    fields: [rid, fileName, status, progress, sttResults, custom]
    note: Also created transiently by post-stt-v1-sync-transcripts, which returns the result inline and issues no durable rid.
  - name: SttResult
    parent: TranscriptionJob
    fields: [transcript, words, keywords, speaker, sentiment]
  - name: Word
    parent: SttResult
    fields: [word, startTime, endTime, speaker]
    note: >-
      startTime/endTime are protobuf-Duration-shaped objects ({seconds, nanos}) rather than floats — a
      Google-STT-shaped detail carried into the REST surface. The guide's example responses also show a
      segmentId on each word; that field appears in the published examples but not in the OpenAPI schema.
  - name: MinutesJob
    id_field: rid
    created_by: [post-nlp-v1-async-minutes]
    read_by: [get-nlp-v1-async-minutes-rid]
    lifecycle_field: status
  - name: SummaryResult
    created_by: [post-nlp-v1-sync-summaries]
    persistent: false
  - name: ParagraphResult
    created_by: [post-nlp-v1-sync-paragraphs]
    persistent: false
  - name: SynthesizedAudio
    created_by: [post-tts-v1-sync-audios]
    persistent: false
    media_type: audio/wav
  - name: SttConfig
    kind: value object
    embedded_in: [TranscriptionJob request]
    fields: [model, language, keywordBoost, multiChannel, speakerDiarization]
  - name: NlpConfig
    kind: value object
    embedded_in: [TranscriptionJob request]
    fields: [sentimentAnalysis, keywordExtraction, summary]
  - name: Callback
    kind: value object
    embedded_in: [TranscriptionJob request, MinutesJob request]
    fields: [url, headers]
  - name: StreamingSession
    kind: transient
    contract: grpc/actionpower-speech.proto
    fields: [language_code, interim_results, max_silence_ms]
    emits: [StreamingRecognitionResult]
    note: gRPC only; no rid, no persistence, nothing to poll.
relationships:
  - {from: TranscriptionJob, to: SttResult, kind: has_many, via: sttResults}
  - {from: SttResult, to: Word, kind: has_many, via: words}
  - {from: TranscriptionJob, to: SttConfig, kind: has_one, via: sttConfig}
  - {from: TranscriptionJob, to: NlpConfig, kind: has_one, via: nlpConfig}
  - {from: TranscriptionJob, to: Callback, kind: has_one, via: callback}
  - {from: MinutesJob, to: Callback, kind: has_one, via: callback}
  - {from: StreamingSession, to: StreamingRecognitionResult, kind: has_many, via: response stream}
identifiers:
  - name: rid
    format: UUID
    prefix: none
    scope: account
    note: >-
      No type prefix, so a transcription rid and a minutes rid are indistinguishable by inspection —
      the caller must remember which endpoint issued it, because polling the wrong one returns 404.
gaps:
  - components.schemas is empty; every schema is inline, so nothing is reusable by $ref and a generator emits duplicate types per operation.
  - No collection, update or delete operations exist for any entity, so a caller cannot enumerate or remove past jobs through the API.
  - Retention of transcripts and uploaded media on the provider side is not stated in the contract.