generated: '2026-08-26' method: searched source: https://docs.plamo.preferredai.jp/en/api name: Preferred Networks PLaMo API conformance description: >- Cross-cutting and domain-standard conformance assertions for the PLaMo API. Preferred Networks publishes no OpenAPI, so nothing here is derived from a machine-readable contract; each entry is evidenced against a specific statement or observed response on the provider's own surface. Absence is recorded as conforms:false with the probe that established it. standards: - id: openai-chat-completions name: OpenAI Chat Completions API (de facto LLM inference interface) conforms: true level: compatible-with-documented-deviations evidence: >- The provider states the interface is compatible with the OpenAI API and that existing code and LLM libraries (openai-python, LangChain) continue to work by pointing base_url at https://api.platform.preferredai.jp/v1. The published surface — POST /v1/chat/completions with messages/model/temperature/top_p/stop/stream/tools/tool_choice/response_format, GET /v1/models, GET /v1/models/{model} — matches the OpenAI shape, and the documented response objects (chat.completion, chat.completion.chunk, model, list) use OpenAI's object names. deviations: - n is restricted to 1 or 2 - logprobs and stop_reason are returned but "currently not supported" - reasoning / reasoning_content / reasoning_effort are PLaMo extensions - POST /v1/tokenize has no OpenAI equivalent - no embeddings, images, audio, files, batch, assistants or responses endpoints are published source: https://docs.plamo.preferredai.jp/en/api - id: sse name: Server-Sent Events (streaming transport) conforms: true evidence: >- Streaming responses are documented as server-sent events terminating with "data: [DONE]", with per-chunk usage available via stream_options.include_usage. source: https://docs.plamo.preferredai.jp/en/api - id: json-schema name: JSON Schema (structured outputs and function parameters) conforms: true evidence: >- response_format {"type":"json_schema","json_schema":{...}} constrains generation to a caller-supplied JSON Schema, with a strict flag; tool function parameters are likewise declared in JSON Schema, and the reference links to json-schema.org. caveat: '"Supported JSON Schema features are limited" — the supported subset is not enumerated.' source: https://docs.plamo.preferredai.jp/en/api - id: mcp name: Model Context Protocol conforms: true scope: PLaMo Translate CLI only, not the PLaMo API evidence: >- `plamo-translate server` starts an MCP server that MCP clients reach at http://localhost:8000/mcp; the package depends on mcp[cli] and ships a show-claude-config command emitting an mcpServers block. The hosted PLaMo API exposes no MCP endpoint. source: https://github.com/pfnet/plamo-translate-cli - id: oauth2 name: OAuth 2.0 conforms: false evidence: >- Authentication is a static bearer API key issued from the console. No authorization server, token endpoint or scope model is documented, and /.well-known/oauth-authorization-server returns 404 on api.platform.preferredai.jp. - id: oidc name: OpenID Connect conforms: false evidence: /.well-known/openid-configuration returns HTTP 404 on every PFN and PLaMo host probed. - id: rfc9457 name: RFC 9457 Problem Details for HTTP APIs conforms: false evidence: >- Errors are returned as {"message":"..."} with content-type application/json, not application/problem+json — observed on an unauthenticated request to /v1/models (HTTP 400) and on /openapi.json (HTTP 404). - id: rfc8594 name: RFC 8594 Sunset header / deprecation signalling conforms: false evidence: >- Model and endpoint retirements are announced with dates in prose, but no Sunset or Deprecation response header is documented. See lifecycle/preferred-networks-lifecycle.yml. - id: ratelimit-headers name: IETF RateLimit header fields conforms: false evidence: >- No RateLimit-*, X-RateLimit-* or Retry-After header is documented; the limitations page states only that the API "returns error" on exhaustion. - id: rfc9116 name: RFC 9116 security.txt conforms: false evidence: /.well-known/security.txt returns HTTP 404 on www.preferred.jp, api.platform.preferredai.jp and docs.plamo.preferredai.jp. - id: pagination name: Documented pagination conforms: false evidence: No paged collection is exposed; /v1/models returns a complete list. - id: idempotency name: Idempotent request replay conforms: false evidence: >- No idempotency key or deduplication window is documented, and seed is explicitly not a determinism guarantee. - id: iso-27001 name: ISO/IEC 27001 conforms: true scope: AI Products and Solutions Division (development, commissioned work, provision of products and services) evidence: >- Certification stated in the corporate site footer and on the AI Governance page. Certificate number, registrar and whether the PLaMo API platform sits inside the certified scope are not published. See security/preferred-networks-trust-center.yml. source: https://www.preferred.jp/en/company/aipolicy domain_standard: market: Large language model / generative AI inference APIs assessment: >- This market's de facto interface standard is the OpenAI Chat Completions shape, and the PLaMo API declares conformance to it in its own reference rather than only on a marketing page: the endpoint paths, request parameter names, response object names (chat.completion, chat.completion.chunk, model, list) and the SSE termination sentinel are the OpenAI ones, and the documented integration path is to repoint an existing OpenAI or LangChain client at the PLaMo base URL. A buyer already speaking that interface integrates with no bespoke connector. standard_id: openai-chat-completions conforms: true confidence: high evidence_location: >- https://docs.plamo.preferredai.jp/en/api — endpoint definitions, request-parameter tables and object list; https://docs.plamo.preferredai.jp/en/getting-started — openai and langchain-openai integration examples setting base_url. caveat: >- There is no formal specification body or conformance suite for this de facto standard, and PFN publishes no machine-readable contract, so the claim is verified against documentation and example payloads rather than against a spec.