generated: '2026-08-01'
method: derived
source: openapi/decart-api-openapi-original.json
description: 'Entity-relationship graph derived from the Decart OpenAPI schemas and
  operation paths. The model is deliberately small: Decart is a generation platform,
  not a records system. Four addressable entities exist — Job, File, ClientToken and
  Model — plus a RealtimeSession that is observable through quota and token constraints
  but is never addressed by id in the REST contract.'
entities:
- name: Job
  description: An asynchronous video or restyle generation submitted to the Queue
    API. Created by POSTing multipart media plus a prompt to a per-model route, then
    polled to completion and downloaded.
  id_field: job_id
  id_type: string
  addressed_by: /v1/jobs/{job_id}
  created_by:
  - lucy_2_5_job
  - lucy_2_1_job
  - lucy_latest_job
  - lucy_edit_v2v_job
  - lucy_clip_latest_job
  - lucy_restyle_2_job
  - lucy_restyle_latest_job
  - lucy_vton_3_job
  - lucy_vton_2_job
  - lucy_vton_latest_job
  read_by: [get_job_v1_jobs__job_id__get, get_job_content_v1_jobs__job_id__content_get]
  fields_documented:
    status: Job status; terminal value "completed"
  note: The 200 response of get_job is typed as an empty schema in both published
    specs, so the job object's full shape is not machine-readable from the contract.
    Only the status field is documented in prose.
- name: File
  description: An uploaded asset (typically a reference image) that can be reused
    across generations by id instead of being re-encoded on every call.
  id_field: id
  id_prefix: file_
  id_type: string
  addressed_by: /v1/files/{file_id}
  schema: FileReferenceResponse
  created_by: [upload_file_v1_files_post]
  read_by: [get_file_metadata_v1_files__file_id__get]
  deleted_by: [delete_file_v1_files__file_id__delete]
  fields:
  - {name: id, type: string, required: true}
  - {name: filename, type: string|null, required: true}
  - {name: mime_type, type: string, required: true}
  - {name: size_bytes, type: integer, required: true}
  - {name: created_at, type: date-time, required: true}
  - {name: expires_at, type: date-time|null, required: true}
  lifecycle:
    ttl_seconds: 60 to 2592000, or "persistent"
    default: 24 hours
- name: ClientToken
  description: A short-lived credential minted server-side for browser and mobile
    realtime sessions.
  id_prefix: ek_
  addressed_by: null
  schema: ClientTokenResponse
  created_by: [create_client_token_v1_client_tokens_post]
  request_schema: ClientTokenRequest
  fields:
  - {name: apiKey, type: string, required: true}
  - {name: token, type: string|null, required: false}
  - {name: expiresAt, type: string, required: true}
  - {name: permissions, type: object|null, required: false}
  - {name: constraints, type: object|null, required: false}
  note: Write-only entity — there is no read, list, or revoke operation. A token expires
    rather than being deleted.
- name: Model
  description: A named generation model. Public names include rolling -latest aliases
    that resolve to a canonical model id.
  id_field: model_id
  addressed_by: null
  schema: ModelResolveResponse
  resolved_by: [resolve_model_aliases]
  request_schema: ModelResolveRequest
  note: Models are not a CRUD resource. The only operation is alias resolution; the
    catalog itself is published as documentation and pricing tables, not as an API
    listing.
- name: RealtimeSession
  description: A live WebRTC generation session. Not addressable by id in the REST
    contract — it is created by the SDK over WebRTC and observed in aggregate through
    the quota endpoint.
  addressed_by: null
  observed_by: [get_realtime_quota_v1_realtime_quota_get]
  schema: RealtimeQuotaResponse
  fields:
  - {name: limit, type: integer|null}
  - {name: active, type: integer|null}
  - {name: remaining, type: integer|null}
  constrained_by: TokenConstraints.realtime.maxSessionDuration
relationships:
- from: Job
  to: File
  type: has_many
  via: reference_image / data submitted by file_ id
  confidence: medium
  evidence: The Files API documents passing a file_ id by reference to job submission
    and realtime connect/set/setImage instead of re-uploading the asset. The multipart
    body schemas type `data` and `reference_image` as binary, so the by-reference form
    is documented in prose rather than in the spec.
- from: Job
  to: Model
  type: belongs_to
  via: route path segment (/v1/jobs/<model>)
  confidence: high
  evidence: The model is encoded in the path, one route per model id or alias, rather
    than as a request field.
- from: ClientToken
  to: Model
  type: has_many
  via: allowedModels
  confidence: high
  evidence: ClientTokenRequest.allowedModels restricts which models the token may
    access (max 20).
- from: ClientToken
  to: RealtimeSession
  type: has_many
  via: constraints.realtime.maxSessionDuration
  confidence: high
  evidence: TokenConstraints.realtime caps the duration of sessions opened with the
    token.
- from: RealtimeSession
  to: File
  type: has_many
  via: file_ reference images passed to connect/set/setImage
  confidence: medium
  evidence: May 2026 changelog — Files API for realtime reference images.
grpc_session_model:
  service: decart.inference.a2v.A2VService
  entity: A2V session
  id_field: session_id
  note: session_id is a token, not bound to the TCP connection; a dropped gRPC channel
    can be reattached with the same session_id. Lifecycle is Initialize → Prompt →
    Infer (loop) → Finish.
  source: grpc/decart-a2v.proto
render: null