Reference implementation · VCF Command Center

A2A Enterprise Agent Plane

A scenario-driven reference implementation of governed AI operations across VMware Cloud Foundation. It correlates evidence, checks policy, routes tools, and holds risky actions for approval, keeping a replayable decision trail throughout.

5 Control-plane layers

13 API endpoints

2 Deployment profiles

Python FastAPI · Docker

Open repo Setup & usage guide

What it does

the operator loop

The demo seeds a VCF-style estate with workload domains, protected services, and integration fabric health, then runs curated executive scenarios for performance, security, and capacity incidents.

Correlates evidence

Pulls simulated signals from VCF Operations, Networks, Logs, SDDC Manager, NSX, Avi, and Automation into a single incident view.

Plans governed action

Generates an action plan carrying confidence, blast radius, and rollback hints, so the operator sees the cost of being wrong before acting.

Splits by risk

Auto-executes low-risk actions through simulated fulfillment adapters, and holds higher-risk actions at an explicit approval gate.

Verifies and replays

Confirms outcomes after execution and keeps a replayable decision trail, so every step can be audited after the fact.

The control-plane pattern

five layers

The repo is the runnable version of the control-plane pattern discussed on the blog: autonomy is bounded by policy and identity, not by hope.

Application

Executive command center and a scenario-driven operator workflow.

Agent control plane

Evidence correlation, policy checks, tool routing, approvals, replay, and audit.

Security

Explicit approval gates and governed execution paths.

Runtime

API service, simulated VCF adapters, and session lifecycle.

Infrastructure

Profile-based deployment: local for labs, vcf for a stricter production posture.

Quick start

runs locally

Self-hosted. Clone the repo, install dependencies, and launch the API service; the command center is served at the root and the API docs at /docs.

# Windows PowerShell python -m venv .venv .\.venv\Scripts\Activate.ps1 pip install -r requirements.txt $env:A2A_CONFIG = "config/local.yaml" python -m uvicorn app.main:app --host 0.0.0.0 --port 9000 --reload # then open http://localhost:9000/ # or run it with Docker: docker compose up --build

Switch A2A_CONFIG to config/vcf.yaml for the stricter profile, which tightens policy and keeps the estate in a production posture.

API surface

FastAPI

GET /api/dashboard GET /api/scenarios POST /api/scenarios/{id}/activate GET /api/sessions GET /api/sessions/{id} POST /api/sessions/{id}/verify POST /api/demo/reset POST /api/incidents POST /api/incidents/{id}/triage GET /api/actions POST /api/actions/{id}/approve GET /api/audit GET /api/overview

Read this before you demo it. The project is platform-agnostic by design and ships with mock adapters, so fulfillment is simulated rather than live. The vSphere adapter (pyvmomi) is a deliberate stub. To connect a nested VCF lab, replace the mock implementations with real vSphere, Kubernetes, and ITSM integrations and use config/vcf.yaml as the baseline.

Open repo All apps Connect on LinkedIn

Views expressed here are my own and do not represent the views of any current or former employer, client, or affiliated organization.