generated: '2026-06-20'
method: derived
source: >-
  openapi/*-openapi.yml; llms.txt (build.nvidia.com); docs.nvidia.com/nim
notes: >-
  DERIVED from the repo OpenAPI plus the provider's own OpenAI-compatibility claims in
  build.nvidia.com/llms.txt ("All NIM model endpoints implement the OpenAI Chat Completions
  API"). NIM is an inference API secured by a Bearer API key; it does not implement OAuth2,
  OIDC, FHIR, FAPI, SCIM, OData, PSD2, or RFC 9457 problem+json.
standards:
  - id: openai-chat-completions
    conforms: true
    evidence: >-
      /v1/chat/completions, /v1/completions, /v1/embeddings, /v1/models mirror the OpenAI
      request/response schemas; provider states OpenAI SDK compatibility in llms.txt.
  - id: openapi-3.1
    conforms: true
    evidence: all specs declare openapi 3.1.0
  - id: sse-streaming
    conforms: true
    evidence: chat/completions expose text/event-stream responses terminated by data:[DONE]
  - id: grpc
    conforms: true
    evidence: Speech (Riva) NIMs publish proto3 services (grpc/nvidia-nim-riva_*.proto)
  - id: prometheus-metrics
    conforms: true
    evidence: self-hosted NIM containers expose /v1/metrics (health-api spec)
  - id: bearer-token-auth
    conforms: true
    evidence: securitySchemes BearerAuth (http bearer, nvapi-... key) across all specs
  - id: oauth2
    conforms: false
  - id: oidc
    conforms: false
  - id: rfc9457-problem-details
    conforms: false
    evidence: error responses use plain descriptions, not application/problem+json
  - id: json-api
    conforms: false
  - id: fhir-r4
    conforms: false
  - id: fapi
    conforms: false
  - id: scim
    conforms: false
  - id: odata
    conforms: false
  - id: cursor-pagination
    conforms: false
    evidence: inference endpoints are stateless; no list pagination beyond /v1/models