Flagship product

Agentic Command Center

A secure control layer for developers building agentic AI orchestration, automation and workflow systems — and a command surface for the enterprise teams accountable for them.

Live surface

The console your operators actually live in.

metzu // command-center
LIVE · tenant=acme

Active agents

14

of 27 registered

Token spend (today)

$1284.40

budget $5,000

Pending approvals

3

2 high-risk

Agent registry
  • wire-transfer-approver

    AGT-0421 · core-banking.transfer

    escalated
  • kyc-onboarding-bot

    AGT-0418 · idv.verify

    active
  • claims-triage

    AGT-0411 · crm.ticket.create

    active
  • outbound-campaign

    AGT-0407 · voice.dial

    paused
  • data-export-helper

    AGT-0398 · warehouse.query

    blocked
Live trace
  • 14:02:11POLICYwire-transfer-approver → human approval required (>$10k)
  • 14:02:09TRACEkyc-onboarding-bot ▸ idv.verify ▸ 312ms
  • 14:02:07BLOCKdata-export-helper denied: tool warehouse.query not allowed
  • 14:02:04COSTtenant=acme department=ops spend +$1.84
  • 14:02:01TRACEclaims-triage ▸ crm.ticket.create ▸ 198ms

Core capabilities

Six capabilities — six places agents go wrong without you.

CapabilityDescriptionEnterprise value
Agent Policy ControlDefine what each AI agent can access, execute, approve, escalate or reject.Reduces unauthorized or unsafe agent behavior.
Tool Usage GovernanceControl APIs, plugins, files, databases, business systems and external services.Prevents uncontrolled tool execution and data exposure.
Workflow TracingTrack multi-step automation, handoffs, retries, failures and agent-to-agent activity.Improves debugging, auditability and operational visibility.
Human Approval WorkflowsRequire approval before sensitive actions like payments, external messages, exports or production changes.Keeps high-risk workflows under enterprise control.
Audit TrailCapture prompt summaries, model responses, tool calls, user actions, decisions and system events.Supports compliance, incident review and accountability.
Cost ControlsTrack and limit LLM usage by user, agent, app, workflow, model or department.Controls token spend and prevents runaway automation cost.

Integration flow

Five steps to ship an accountable agent.

  1. STEP 01

    Register Agent

    Add each AI agent, workflow, automation task and tool to the MetzuAI registry.

  2. STEP 02

    Attach Policy

    Assign permissions, model limits, allowed tools, escalation rules and approval gates.

  3. STEP 03

    Trace Execution

    Capture prompts, responses, tool calls, token usage, latency and workflow status.

  4. STEP 04

    Control Actions

    Allow, deny, pause, resume or stop actions based on role, policy and risk level.

  5. STEP 05

    Report & Optimize

    Analyze performance, cost, compliance, exceptions and optimization opportunities.

SDK

Wire it into your agents in minutes.

import { Metzu } from "@metzuai/sdk";

const metzu = new Metzu({ apiKey: process.env.METZU_KEY });

// 1. Register
await metzu.agents.register({
  id: "wire-transfer-approver",
  owner: "ops@bank.com",
  env: "production",
});

// 2. Policy
await metzu.policy.attach("wire-transfer-approver", {
  tools: { allow: ["core-banking.transfer", "crm.lookup"] },
  models: { allow: ["gpt-4o", "claude-3-5-sonnet"], maxTokens: 8000 },
  approvals: [{ when: "amount > 10000", role: "ops.lead" }],
});

// 3. Trace every step
const run = metzu.trace("wire-transfer-approver");
await run.tool("core-banking.transfer", { amount, to });

Observability · Command · Cost

Three operating modes, one control plane.

OBSERVABILITY

Trace · Debug · Measure · Audit

Agent steps, workflow paths, tool calls and system events with latency, success rate, token usage and full audit trails.

COMMAND

Registry · Status · Kill switch

Pause, resume, escalate or stop any agent, workflow, tool or automation instantly. Action-level controls per role.

COST

Usage · Budgets · Models · Alerts

Track tokens by user, agent, app, workflow, model or department. Set budgets and trigger alerts on anomalies.

RBAC

One control surface, six different jobs.

RoleAccessPurpose
AdminFull access to agents, policies, users, workflows, logs, cost controls and environments.Owns platform configuration and governance.
DeveloperIntegrate the SDK, create agents, inspect logs, debug workflows and view tool traces.Builds and maintains agentic applications.
Compliance OfficerAudit logs, policy history, approval records and exception reports.Reviews risk, control and accountability evidence.
Finance TeamToken usage, model spend, budgets, alerts and cost optimization reports.Controls LLM consumption and budget discipline.
OperatorPause, resume, stop, escalate or approve workflows based on permissions.Runs day-to-day command and control operations.
Business UserApproved dashboards, workflow results, summaries and operational reports.Uses AI outcomes without platform administration.

Where it runs

Developer use cases & supported environments.

Developer use cases

  • · Agentic AI orchestration platforms
  • · Enterprise AI assistants
  • · Multi-agent systems
  • · Workflow automation engines
  • · API-driven agents
  • · Internal copilots and operators

Supported environments

  • · Backend services
  • · Agent frameworks
  • · Automation platforms
  • · Enterprise apps
  • · Cloud
  • · Private cloud
  • · On-premise

Ready to put your agents under control?

Book a 30-minute walkthrough of the MetzuAI SDK — governance, command and cost control, live on your stack.

Book a demo