APP · 04 — THE ORCHESTRATOR

Nexus — the orchestrator.

Nexus is the MCP Host of the cognitive server. It carries the approved Skill Catalog, four granular RBAC roles — admin, director, operator, viewer — and the compliance posture surface that turns every reasoning step into evidence. Bridge transports the JSON-RPC traffic, Shield evaluates the scope, Chain seals the result, and a Firebase mirror keeps the live posture visible across the operator's regulators, auditors and lines of business. Where Vault remembers and Hub connects, Nexus decides who is allowed to do what — and proves it.

Modules

What ships inside Nexus

MOD-01

MCP Host

The sovereign Model Context Protocol host of the appliance. Routes every Tool, Resource and Prompt call through scope evaluation, tenant binding and Chain sealing before the model touches the result.

MOD-02

Skill Catalog

Registry of approved, versioned skills — incident triage, conformity assessment, vendor due diligence, KPI extraction. Each skill is signed by Shield, replayable from Chain and graduates from experimental in Vault to corporate in Nexus.

MOD-03

RBAC & Compliance

Four granular roles (admin, director, operator, viewer) with scope-by-scope policy bindings. Every ALLOWED and DENIED call is appended to the compliance log with regulator-aligned tags (NIS2, DORA, EU AI Act).

MOD-04

Firebase Posture Mirror

Live, read-only mirror of the appliance's compliance posture into a Firebase project the operator owns. Multi-device dashboards, regulator views and incident channels — without a raw byte of operator data leaving the perimeter.

Hooks

Programmatic surface

useMcpHost

useMcpHost() => { invokeTool, readResource, runPrompt, lastTrace }
Programmatic surface for the sovereign MCP Host. Every call is scope-checked by Shield, signed by Chain, and returns alongside the trace identifier it generated.

useSkillCatalog

useSkillCatalog(status?: SkillStatus) => { skills, promote, deprecate }
Reads the approved Skill Catalog. promote() graduates an experimental skill from Vault into the corporate catalog after policy review; deprecate() retires one with a regulatory reason code.

useRbac

useRbac() => { role, scopes, can(action), evaluateBatch(actions) }
Reads the current operator's role and active scope set, and offers a can() predicate that mirrors the server-side policy without round-tripping for every UI decision.

useComplianceLog

useComplianceLog(filters?: ComplianceFilters) => { entries, export, regulatorView }
Streams ALLOWED/DENIED entries from the compliance ledger with regulator-aligned filtering (NIS2, DORA, EU AI Act). export() produces a signed, replayable evidence pack.

Integration example

Drop Nexus into a tenant app

// Run an EU AI Act conformity assessment through the sovereign MCP host
import {
  useMcpHost,
  useRbac,
} from "@cognitivserver/nexus";

export function ConformityRunner({ systemId }: { systemId: string }) {
  const { runPrompt, lastTrace } = useMcpHost();
  const { can } = useRbac();

  async function run() {
    if (!can("compliance:run")) throw new Error("Scope denied by Shield");
    const result = await runPrompt({
      name: "ai_act_conformity",
      args: { systemId, regime: "EU_AI_ACT" },
    });
    // The Chain trace is the evidence — keep it next to the result.
    return { result, traceId: lastTrace?.id };
  }

  return <RunButton onRun={run} />;
}

Language · ts

Engage

Want to see this in your data center?

Request a briefing →