NueOS Documentation
ArchitectureManaged Inference Relay

Managed Inference Relay

Overview of NueOS Cloud managed model routing and regional inference relay architecture

Managed Inference Relay

The managed inference relay is the NueOS Cloud path for tenant-authenticated model inference where Nue chooses and operates the backend model connector.

Instead of asking clients to configure Vertex AI, AWS Bedrock, Azure AI Foundry, direct frontier APIs, or hosted open-weight pools, clients call a first-party Nue endpoint. Nue authenticates the tenant, selects the backend route, forwards the request through a low-overhead regional relay, and records usage for quota enforcement, metering, and operations.

NueOS client

nue-ai-api / nue-cloud provider leaf

Nue Cloud control plane

regional inference relay

server-side provider connector

Vertex / Bedrock / Azure / OpenAI / Anthropic / vLLM / etc.

Why this exists

The current provider leaf model works well for direct or BYOK providers with static API keys and stable endpoints. Cloud model platforms have a different shape: project/account identity, regional endpoints, OAuth/IAM/SigV4 authentication, quotas, capacity, tenant policy, and billing all matter.

The managed relay keeps those concerns server-side while presenting a stable Nue-facing API to clients.

Reading order

  1. Architecture Overview — problem statement, goals, non-goals, and core concepts.
  2. Control Plane and Routing — tenant auth, model aliases, route decisions, sessions, and relay tokens.
  3. Regional Relay Data Plane — low-overhead streaming relay behavior and runtime responsibilities.
  4. Server-side Provider Connectors — Vertex, Bedrock, Azure, direct API, and open-weight connector responsibilities.
  5. Usage Metering and Observability — usage ledger, usage accounting, quota, traces, and privacy constraints.
  6. Implementation Plan — phased rollout from a static backend to managed routing, metering, and cost controls.

Relation to provider adapters

Provider adapters remain the right path for direct provider integrations such as OpenAI, Anthropic, Groq, Mistral, Together AI, Gemini API, and local runtimes.

The managed inference relay is a different path for NueOS Cloud subscriptions where Nue manages provider credentials, routing, quota, billing, fallback, and provider selection.

On this page