generated: '2026-09-02'
method: derived
source: openapi/standard-compute-openapi.json
also:
  - https://standardcompute.com/getting-started
  - https://standardcompute.com/security
  - probed: GET https://api.stdcmpt.com/v1/models (2026-09-02)
standards:
  - id: openapi
    name: OpenAPI Specification
    conforms: true
    version: 3.1.0
    evidence: >-
      https://api.stdcmpt.com/openapi.json returns a parsing OpenAPI 3.1.0 document
      with nine operations (HTTP 200, application/json, probed 2026-09-02). Saved
      verbatim to openapi/standard-compute-openapi.json.
  - id: oauth2
    conforms: false
    evidence: >-
      No oauth2 securityScheme in the spec, no OAuth documentation, and
      /.well-known/oauth-authorization-server 404s on both hosts. Authentication is
      a single Bearer API key.
  - id: oidc
    conforms: false
    evidence: /.well-known/openid-configuration returns 404 on both hosts.
  - id: rfc9457
    name: Problem Details for HTTP APIs
    conforms: false
    evidence: >-
      Errors use the OpenAI envelope {"error":{"message","type"}} with content-type
      application/json, observed live on a 401. No application/problem+json.
  - id: rfc9116
    name: security.txt
    conforms: false
    evidence: >-
      /.well-known/security.txt returns 404 on both hosts, although a responsible
      disclosure policy IS published at https://standardcompute.com/security.
  - id: rfc8594
    name: Sunset header
    conforms: false
    evidence: No Sunset or Deprecation header is documented or observed.
  - id: idempotency
    conforms: false
    evidence: No idempotency key documented; see conventions/standard-compute-conventions.yml.
  - id: pagination
    conforms: false
    not_applicable: true
    evidence: No collection endpoint paginates; GET /v1/models returns the full pool unpaged.
  - id: json-api
    conforms: false
    evidence: Media type is application/json, not application/vnd.api+json.
  - id: sse
    name: Server-Sent Events streaming
    conforms: true
    evidence: >-
      Inherited from the OpenAI and Anthropic wire formats the gateway implements;
      documented by the setup guides for streaming agents. NOT described in the
      published OpenAPI, whose response schemas are empty.
    confidence: medium
domain_standards:
  - id: openai-chat-completions
    name: OpenAI Chat Completions / Models API (de-facto LLM inference interface)
    conforms: true
    market: LLM inference and AI gateways
    evidence: >-
      Declared by the CONTRACT, not by a marketing page: the published OpenAPI
      defines POST /v1/chat/completions (operationId
      chat_completions_v1_chat_completions_post), POST /v1/completions, POST
      /v1/responses and GET /v1/models at the OpenAI path shapes, and the live
      GET /v1/models response is the OpenAI list envelope
      {"object":"list","data":[{"id","object","created","owned_by",...}]}.
    signature:
      spec_location: openapi/standard-compute-openapi.json#/paths/~1v1~1chat~1completions
      probe: GET https://api.stdcmpt.com/v1/models -> 200, OpenAI list envelope
    buyer_impact: >-
      This is the whole product thesis and it is the reason the integration cost is
      a one-line base-URL change. Any tool already speaking the OpenAI wire format
      integrates with no bespoke connector; the provider ships no SDK precisely
      because the standard's SDKs already work.
  - id: anthropic-messages
    name: Anthropic Messages API (second de-facto interface for coding agents)
    conforms: true
    market: LLM inference for agentic coding tools
    evidence: >-
      POST /v1/messages (operationId anthropic_messages_v1_messages_post) and POST
      /v1/messages/count_tokens (anthropic_count_tokens_v1_messages_count_tokens_post)
      are declared in the published OpenAPI. GET /v1/models carries an
      anthropic_family_tier field per model (opus / sonnet / fable), confirming the
      Anthropic family model is a first-class concept in the routing layer and not
      just a compatibility shim.
    signature:
      spec_location: openapi/standard-compute-openapi.json#/paths/~1v1~1messages
      probe: GET https://api.stdcmpt.com/v1/models -> anthropic_family_tier present on every entry
    buyer_impact: >-
      Supporting BOTH wire formats behind one key is uncommon; it is what lets Claude
      Code run against this gateway with only ANTHROPIC_BASE_URL changed.
    note: >-
      The Anthropic path deliberately takes a base URL WITHOUT the /v1 suffix
      because the client appends /v1/messages itself.
compliance_certifications:
  published: false
  claims: []
  note: >-
    Recorded as an honest absence, and the provider is unusually explicit about it.
    The security page states that certifications vary by upstream provider — "not
    every provider in our pool holds the same third-party certifications (e.g. SOC
    2)" — and directs buyers with compliance requirements to the upstream provider's
    own terms rather than relying on Standard Compute's summary. No SOC 2, ISO 27001,
    PCI, HIPAA or FedRAMP claim is made for Standard Compute itself, and no trust
    center exists. NO Compliance or TrustCenter pointer is wired.
  source: https://standardcompute.com/security
regulatory_regime: none mapped (independent LLM inference vendor; no sector regime applies)