Wolfberg WOLFBERG

TL;DR

Keystone is an AI employee operating system — six modules, seven AI roles, any service vertical. It runs six AI employees in production today across two live operator instances. I built it solo in 41 hours for $40 in API costs.

Currently Building

KEYSTONE.

Operational infrastructure for AI-native businesses.

Not a product for one vertical. Operational infrastructure for all of them. Built on AWS, serverless, multi-tenant, designed to scale into billions of revenue. Already running its first vertical end-to-end.

The Thesis

Every service vertical has the same operational structure: intake, triage, dispatch, communication, collection, renewal, escalation.

Keystone handles all of it — AI-native, serverless, at near-zero unit cost — so operators don't have to build it.

One platform. Many verticals. One architecture.

6 / 6

Modules in production

6 / 7

AI employees live

NIST 800-171

Security hardening

2 live

Operator instances

The Platform

Six modules. One spine.

Each module owns a single concern. Together they form the operational substrate for any service business.

Arch

Production

Workflow Engine

EventBridge, SQS, Step Functions orchestration. Multi-stage state machines with audit trails, dead-letter queues, and waitForTaskToken — no polling.

Lintel

Production

AI Employee Runtime

Where every AI employee lives. Claude (Bedrock) agent loop, tool use, conversation state, system prompts. Every role-specific employee shares one runtime.

Sanctum

Production

Data & API

JWT-verified API Lambda, role-scoped DynamoDB, KMS-encrypted S3, Cognito PKCE OAuth2, full audit trail, multi-tenant isolation. NIST 800-171 hardened, MFA enforced.

Buttress

Production

Notification & Escalation

SES email, Twilio SMS, Twilio voice with Polly Generative — AI-personalized emergency calls. Escalation retry loop with deterministic state machine.

Plinth

Production

Per-Tenant Configuration

SSM Parameter Store. Add a new operator = create SSM params. No code change. No redeployment. White-label branding per operator.

Portico

Production

Frontend SDK

React + TypeScript role-aware shell. Tenant, owner, vendor, admin views from the same codebase. White-label from Plinth config. Deployed via CloudFront.

AI Employees

Real interfaces. Real models. Real data.

Each AI employee is a Claude agent with role-scoped data access and tool use. They answer questions from live data, generate reports and documents on demand, and escalate to a human operator when they should. Available on web and mobile, same conversation thread, same context.

AE-00
Steward
Tenant reports an issue → AI triages, gets approval, dispatches the right vendor
Live
AE-01
Concierge
New contact → classified and routed to the right next step
Live
AE-02
Liaison
Owner asks anything about their portfolio → AI answers from real data, generates reports
Live
AE-03
Custodian
Lease expiring → 90 / 60 / 30-day outreach with conversation capability
Live
AE-04
Bursar
Rent late → reminder sequences and payment conversations within defined limits
Live
AE-05
Quartermaster
Vendor assigned → AI answers questions, tracks the work, captures completion
Live
AE-06
Envoy
Prospective tenant asks about a unit → AI answers and pre-qualifies
Sprint 5

Use Case 1

Wolfberg Property Management.

The setup

A real property management business in Northern Virginia. AI-native from day one. Built end-to-end on Keystone — every workflow, every conversation, every escalation runs through the platform.

What's running

Maintenance triage with multi-turn AI conversation. Emergency voice escalation via Twilio. Owner portal with maintenance timeline. Vendor dispatch. Document vault. Multi-tenant infrastructure with operator scoping.

What it proves

Keystone runs a real operating company. Not a demo. Not a prototype. Production AWS infrastructure handling production workloads — and ready to drop into any other service vertical via a domain pack.

The Deck

Keystone v18.

25 slides. The full platform thesis — architecture, AI workforce, market opportunity, competitive position, use case, and licensing model.

Download PDF ↓

Keystone v18 · Wolfberg LLC Proprietary & Confidential · Updated May 2026

Beyond Use Case 1

The same architecture
runs any service business.

A domain pack defines the stakeholder roles, data schema, tools, and system prompts. Keystone provides the agent runtime, auth and scoping, conversation state, chat UI, outbound triggers, and escalation.

Property management is the first domain pack. Legal services, home services, healthcare practices, accounting, property leasing, government program offices — same infrastructure, different domain pack.

For Operators

Want to run your operation on Keystone?

Tell us about your vertical. We'll tell you what a domain pack looks like.

Start the conversation

If your cloud bill scales linearly with your customer count — we should talk.

Work with me →