generated: '2026-08-04' method: searched source: https://llmboost.mangoboost.io/docs/features/openai-api name: MangoBoost standards conformance description: >- Cross-cutting and industry standards MangoBoost's public surfaces claim or demonstrably implement. Every `conforms: true` row carries a URL that documents the claim. No security or privacy certification (SOC 2, ISO 27001, PCI, HIPAA, FedRAMP) is published anywhere on the MangoBoost surface, so no Compliance pointer is wired into apis.yml. standards: - id: openai-api name: OpenAI REST API compatibility conforms: true scope: Mango LLMBoost Inference Server evidence: >- "LLMBoost serves the OpenAI REST API on /v1. If your app already talks to OpenAI, migration is a one-line base-URL change — no SDK swap, no rewrite." Documented endpoints: /v1/chat/completions, /v1/completions, /v1/embeddings, /v1/responses, /v1/audio/transcriptions, /v1/audio/translations, /v1/models. url: https://llmboost.mangoboost.io/docs/features/openai-api - id: openai-streaming-sse name: OpenAI streaming (server-sent chunk deltas) conforms: true scope: Mango LLMBoost Inference Server evidence: '`"stream": true` returns chunked deltas via the standard OpenAI client iterator.' url: https://llmboost.mangoboost.io/docs/features/streaming-structured-tools - id: json-schema-structured-output name: JSON Schema constrained decoding (response_format json_schema / json_object) conforms: true scope: Mango LLMBoost Inference Server evidence: >- "Force valid JSON — guided decoding constrains the model to your schema." Both {"type":"json_schema"} and {"type":"json_object"} are supported. url: https://llmboost.mangoboost.io/docs/features/streaming-structured-tools - id: openai-tool-calling name: OpenAI tool / function calling conforms: partial scope: Mango LLMBoost Inference Server evidence: >- Tool schemas are forwarded and tool_choice is honored, but emission "needs a tool-trained instruct model and the matching tool-call parser enabled on the server." url: https://llmboost.mangoboost.io/docs/features/streaming-structured-tools - id: vllm-cli name: vLLM-compatible serve interface conforms: true scope: llmboost serve evidence: >- "llmboost serve exposes a vLLM-compatible serve interface — the same flag names and semantics — so your existing launch scripts, configs, and tooling run against it unchanged." url: https://llmboost.mangoboost.io/docs/running/configuration - id: prometheus-metrics name: Prometheus metrics exposition conforms: true scope: Mango LLMBoost Inference Server evidence: 'GET /metrics is documented as "Prometheus metrics"; the Helm chart bundles kube-prometheus-stack.' url: https://llmboost.mangoboost.io/docs/features/openai-api - id: opi name: Open Programmable Infrastructure (OPI) storage API conforms: true scope: Mango OPI Storage Bridge evidence: >- "MangoBoost provides a gRPC bridge that enables users to interact with the Mango SDK via the OPI gRPC interface." Implements the OPI NVMe frontend/backend calls — CreateNvmeSubsystem, CreateNvmeController, CreateNvmeRemoteController, CreateNvmePath, CreateNvmeNamespace — and is hosted in the Linux Foundation OPI project itself. url: https://github.com/opiproject/opi-mangoboost-bridge - id: grpc name: gRPC conforms: true scope: Mango OPI Storage Bridge evidence: 'Bridge log: "gRPC server listening at [::]:50051", "HTTP Server listening at 8082"; called with grpc_cli.' url: https://sdk.mangoboost.io/docs/guide/docs_opi/guide - id: nvme-of name: NVMe over Fabrics (NVMe/TCP initiator and target) conforms: true scope: Mango BoostX NTI / NTT evidence: >- NVMe-oF subsystems, controllers, namespaces and paths configured with NVMe/TCP transport (trsvcid 4420, NQN nqn.2019-07.io.spdk:cnode0 in the vendor example). url: https://sdk.mangoboost.io/docs/guide/docs_storageboost_ntt/overview - id: spdk name: SPDK (Storage Performance Development Kit) RPC conforms: true scope: Mango DPU storage stack evidence: >- "the OPI bridge translates it into the corresponding MangoBoost SPDK RPC calls"; mango-drivers-spdk is a shipped package; bridge reports SPDK v25.05 firmware revision. url: https://sdk.mangoboost.io/docs/guide/docs_opi/overview - id: rocev2-rdma name: RoCE v2 / RDMA conforms: true scope: Mango BoostX RoCE AI, Mango GPUBoost RDMA evidence: 'Device description "Mango BoostX - RoCE AI"; RDMA PCC API group in the SDK library reference.' url: https://sdk.mangoboost.io/docs/guide/docs_boostx_roce_ai/overview - id: posix name: POSIX API compatibility layer conforms: true scope: Mango SDK (libmango) evidence: '"POSIX emulation APIs to provide compatibility with the standard POSIX APIs."' url: https://sdk.mangoboost.io/docs/api/posix - id: mlperf name: MLPerf Inference / Training / Storage benchmark submissions conforms: true scope: Company-wide evidence: >- "Record MLPerf Inference v5.0, v5.1, and v6.0 results on AMD Instinct GPUs"; MLPerf Training v4.1 and MLPerf Storage v1.0 submissions are announced on the blog. url: https://llmboost.mangoboost.io/docs/benchmarks - id: rfc9457 name: RFC 9457 Problem Details for HTTP APIs conforms: false scope: Mango LLMBoost Inference Server evidence: >- No application/problem+json envelope is documented. LLMBoost inherits the OpenAI error envelope ({"error": {...}}) rather than RFC 9457. - id: oauth2 name: OAuth 2.0 conforms: false scope: All surfaces evidence: >- No OAuth surface. LLMBoost local serving takes no Authorization header at all ("Drop the Authorization header for local serving"); the DPU surfaces are host-local (libmango, gRPC on the card). - id: openapi name: OpenAPI description conforms: false scope: All surfaces evidence: >- No OpenAPI/Swagger document is published at any MangoBoost host. Probed /openapi.json, /openapi.yaml, /swagger.json, /api-docs and /redoc on www., sdk. and llmboost.mangoboost.io — all 404. - id: asyncapi name: AsyncAPI / event surface conforms: false scope: All surfaces evidence: 'No webhooks, event stream or message broker surface is documented. Not applicable to a self-hosted inference server and on-card DPU API.' certifications_published: [] x-evidence: fetched: '2026-08-04' probes: - url: https://llmboost.mangoboost.io/docs/features/openai-api http_status: 200 - url: https://sdk.mangoboost.io/docs/guide/docs_opi/guide http_status: 200 - url: https://www.mangoboost.io/openapi.json http_status: 404