AI/ML & Agents — Nexocloud

Model serving, RAG, guardrails, an MLOps feature store and a federated MCP tool gateway.

The intelligence plane: a model-serving broker with budgets, guardrails, embeddings, knowledge bases and canary promotion; an MLOps plane with a feature store, drift detection, skew checks and serve-guards; and a federated MCP gateway that catalogs tools across nodes so agents call governed, discoverable tools.

Auth is one model across every service here — an SS-01 bearer token plus X-Tenant-Id on every /api/v1 route. Public base https://api.nexocloud.io/api/v1.

3 real backends, 59 operations.

Every path below is a real route from services/<svc>/openapi.json. Each service lists a representative slice of its full contract.

Model Serving SS-15

29 operations

Model serving & RAG — a model broker, embeddings, knowledge bases, guardrails, token budgets and canary promotion.

POST/api/v1/broker/{name}/inferRun inference through the model broker.
POST/api/v1/embeddingsGenerate embeddings.
POST/api/v1/kb/{name}/retrieveRetrieve from a knowledge base (RAG).
POST/api/v1/guardrailsApply input/output guardrails.
POST/api/v1/models/{name}/{version}/canary/advanceAdvance a model canary rollout.

MLOps SS-46

17 operations

MLOps — feature sets, online features, point-in-time lookups, skew checks, drift and serve-guards.

POST/api/v1/feature-setsDefine a feature set.
GET/api/v1/features/{name}/point-in-timePoint-in-time feature lookup for training/serving parity.
GET/api/v1/features/{name}/skew-checkCheck training/serving skew.
POST/api/v1/driftReport a drift observation.
GET/api/v1/models/{name}/{version}/serve-guardCheck whether a model version is safe to serve.

MCP Gateway SS-49

13 operations

Federated MCP gateway — register nodes, sync a federated tool catalog and resolve tool schemas for agents.

POST/api/v1/nodesRegister an MCP node.
GET/api/v1/catalogGet the full federated tool catalog snapshot.
POST/api/v1/catalog/refreshRe-sync tools across all federated nodes.
GET/api/v1/tools/{name}Get a tool's schema and owning node.

A guardrailed RAG answer under budget

Retrieve context from a knowledge base, run inference through the broker under a token budget, and enforce guardrails on the way out — with agents reaching governed tools through the MCP catalog.

RetrievePOST/api/v1/kb/{name}/retrieve

Pull the top-k passages from a tenant knowledge base.

InferPOST/api/v1/broker/{name}/infer

The broker routes to a model under a token budget.

GuardrailPOST/api/v1/guardrails

Screen the output before it reaches the user.

Discover toolsGET/api/v1/catalog

Agents find governed tools through the federated MCP catalog.

Where it sits in the platform.

AI/ML & Agents does not reimplement the platform — it composes it. These are the services it leans on and the seams between them.

Model Serving (SS-15)

The inference broker, embeddings, RAG and guardrails with token budgets.

MLOps (SS-46)

Feature store, skew/drift checks and serve-guards for safe promotion.

MCP Gateway (SS-49)

Federated tool catalog so agents call discoverable, governed tools.

Vault (SS-05) + VRQGO (SS-06)

Model credentials are custodied; every inference is traced and budgeted.