generated: '2026-08-26'
method: derived
source: openapi/multiverse-computing-compactifai-openapi.yml
also_searched:
- https://docs.compactif.ai/openai-compatibility/
- https://docs.compactif.ai/data-privacy/
- https://docs.compactif.ai/faq/
- https://multiversecomputing.com/privacy-policy
- https://multiversecomputing.com/legal-notice
- https://multiversecomputing.com/quality-and-environmental-policy
conformance:
- id: openapi-3.1
  conforms: true
  evidence: 'openapi/_original/multiverse-computing-compactifai-openapi.json declares
    "openapi": "3.1.0" and parses; 13 operations across 12 paths, all with operationIds and
    summaries. Served publicly and unauthenticated at https://api.compactif.ai/openapi.json
    (HTTP 200, application/json), and identically at api-eu.compactif.ai and
    api-us.compactif.ai (byte-identical, sha256 a1ef46ae0e83c94b…).'
- id: oauth2
  conforms: false
  evidence: The only securityScheme is HTTPBearer (http/bearer). No oauth2 flows, no
    /.well-known/oauth-authorization-server (404 on every host), no token endpoint.
- id: oidc
  conforms: false
  evidence: '/.well-known/openid-configuration returns 404 on api.compactif.ai,
    api-eu.compactif.ai, api-us.compactif.ai and docs.compactif.ai (403 S3 AccessDenied),
    and 404 on multiversecomputing.com. End-user sign-in to the Dashboard uses Multiverse
    IAM with an emailed OTP, but no OIDC discovery document is published.'
- id: rfc9457
  conforms: false
  evidence: 'Errors are application/json with an OpenAI-shaped {"error":{...}} envelope, or
    a bare {"detail":...}. No application/problem+json content type appears in the spec or
    on a live 401/404.'
- id: rfc9116-security-txt
  conforms: false
  evidence: /.well-known/security.txt returns 404 on api.compactif.ai, api-eu, api-us and
    multiversecomputing.com, and 403 (S3 AccessDenied) on docs.compactif.ai. Security
    reports are routed through the support form instead.
- id: rfc8594-sunset
  conforms: false
  evidence: No Sunset or Deprecation response header is declared in the spec or observed
    live, and the changelog announces removals after the fact.
- id: pagination
  conforms: true
  evidence: 'GET /v1/batches implements cursor pagination with `after` and `limit` (1–100)
    and returns the OpenAI list envelope BatchList {object, data, has_more, first_id,
    last_id}. Only this one operation paginates.'
- id: idempotency
  conforms: false
  evidence: No Idempotency-Key header on any operation; the string "idempoten" does not
    appear in the spec or in the published documentation. See
    conventions/multiverse-computing-conventions.yml.
- id: server-sent-events
  conforms: true
  evidence: 'Streaming responses are delivered as Server-Sent Events when `stream: true`,
    documented at https://docs.compactif.ai/faq/ and supported on chat completions,
    completions, responses and audio transcriptions.'
- id: json-schema-2020-12
  conforms: true
  evidence: OpenAPI 3.1.0 uses JSON Schema 2020-12 for its component schemas; the spec
    declares 150+ components.schemas and uses anyOf/$ref throughout.
- id: gdpr
  conforms: partial
  evidence: 'Multiverse Computing, S.L. is a Spanish (EU) entity (Paseo Miramon 170,
    Donostia-San Sebastian, tax ID B-75218040, per
    https://multiversecomputing.com/legal-notice) and publishes a privacy policy plus a
    Zero Data Retention statement: prompts and completions are not stored, anonymized
    request logs are deleted after 30 days, and the user identifier is stripped and
    replaced with a random request ID. Dedicated EU (api-eu.compactif.ai) and US
    (api-us.compactif.ai) endpoints allow data-residency selection. No DPA, no
    sub-processor list and no certification is published, so this is recorded as partial
    rather than conformant.'
- id: soc2
  conforms: false
  evidence: No trust center, no SOC 2 report and no certification page found.
    probe-security-programs.py returned trust=none across trust./security./compliance
    paths on multiversecomputing.com and compactif.ai.
- id: iso-27001
  conforms: false
  evidence: Not named anywhere on the website or docs. The only ISO standard mentioned
    company-wide is ISO 14001, and it is described as "certification readiness" (a goal),
    on https://multiversecomputing.com/quality-and-environmental-policy — an environmental
    management standard, not an information-security one, and therefore not an API
    compliance signal.
domain_standard:
  market: LLM inference / model-serving APIs
  standard: OpenAI API compatibility (de-facto interoperability standard for LLM inference
    endpoints)
  declared: true
  conforms: true
  evidence:
  - location: openapi/multiverse-computing-compactifai-openapi.yml paths
    detail: 'The contract implements the OpenAI request/response surface verbatim under
      /v1: /v1/chat/completions, /v1/completions, /v1/responses, /v1/responses/{id},
      /v1/batches (+ list, retrieve, cancel), /v1/files, /v1/files/{id}/content,
      /v1/models, /v1/models/{model}, /v1/audio/transcriptions.'
  - location: openapi operation descriptions
    detail: 'The spec states the compatibility in its own text: "OpenAI-compatible POST
      `/v1/audio/transcriptions`", "Accepts OpenAI ``BatchCreateParams``", "List batch jobs
      (OpenAI-compatible: ``GET /v1/batches`` with ``after``, ``limit``, and list
      envelope)", "Returns a ChatCompletionResponse in OpenAI-compatible format",
      "Matches ``POST /v1/batches/{batch_id}/cancel``".'
  - location: openapi components.schemas
    detail: 'Component names are the OpenAI type names — ChatCompletionRequest,
      ChatCompletionAssistantMessageParam, ChatCompletionContentPartImageParam,
      ResponseOutputMessage, ResponseFunctionToolCall, BatchCreateParams, BatchList,
      FileObject, CompletionUsage — including two schemas whose keys are the literal OpenAI
      Python module paths (openai__types__responses__web_search_tool__UserLocation).'
  - location: https://docs.compactif.ai/openai-compatibility/
    detail: A dedicated documentation page for OpenAI compatibility, linking to
      platform.openai.com API reference pages and to the openai-python and openai-node
      SDKs as the recommended clients.
  buyer_impact: 'A team already calling OpenAI integrates by changing base_url and the
    model ID; no bespoke connector is needed. This is the whole commercial premise of the
    product.'
  caveats:
  - 'Compatibility is a surface claim, not a version claim: no OpenAI API version or dated
    schema revision is named, so drift is unmanaged on both sides.'
  - 'Parameter coverage diverges. ChatCompletionRequest omits `seed` and `logprobs` and
    adds three non-OpenAI parameters — `min_tokens`, `ignore_eos`, `reasoning_enabled` —
    so a strict OpenAI client will not exercise the whole surface and a CompactifAI-tuned
    client is not portable back.'
  - 'The FAQ note that streaming, tools and response_format are supported is documentation,
    not contract: the spec does not mark which models honour which parameter, while the
    changelog shows tool calling arriving model-by-model.'
notes:
- 'REWARD-ONLY reading: LLM inference has no formal SDO standard. OpenAI API compatibility
  is the de-facto one, it is declared inside the contract rather than only on a marketing
  page, and it is implemented across the whole 13-operation surface.'