Three Claude agents
that run property
maintenance end-to-end.

Property managers with 50–500 units waste 15–20 hours a week on manual coordination. EstateOps classifies every ticket, estimates the repair cost, dispatches the right vendor, and keeps the tenant informed — automatically.

See how it works Hire the builder
Claude Sonnet 4.6
Built on Next.js 15
SOC2-ready infra
estateops.ai/tickets/new
Incoming ticket
"HVAC not cooling, indoor unit leaking water onto the floor. Family with 2 kids."
Unit 4B · 2026-04-19 14:32
↓
Agent 1 — Estimator
categoryHVAC · Cooling failure
priorityHIGH
est. cost$380 – $620
responsibilityLANDLORD
vendorAustin HVAC Pros (4.8★)
etatoday 16:00–17:00
How it works

From ticket to tenant comms in under a minute.

A simple three-step pipeline. No dashboards to learn, no rules to configure. Plug it into your existing workflow and let the agents run.

01

Ticket arrives

Tenant submits a maintenance request via portal, SMS, or email. The payload lands in the pipeline with unit context auto-attached.

02

Agents run in sequence

Estimator classifies, prices, and assigns responsibility. Router finds the best vendor by skill, zone, rating. Comms schedules the full message timeline.

03

Humans stay in the loop

Every decision is logged with the agent's reasoning. Override any step, or let the system run autonomously on pre-approved ticket types.

The three agents

Each agent has one clear job.

Small, specialized, auditable. Claude Sonnet 4.6 with tool use — no fragile prompt chains.

Agent 01

Maintenance Estimator

Reads the tenant message, classifies the issue, estimates repair cost using the unit's history, and assigns responsibility — landlord or tenant — with a full rationale.

Claude Sonnet SSE streaming pgvector
Agent 02

Vendor Router

Matches the required skill, service zone, rating, and availability. Calls the vendor database via tool use and returns the top three candidates with reasoning.

Tool use Supabase Geo match
Agent 03

Tenant Comms

Orchestrates the full message timeline: immediate confirmation, T–30min reminder, T+2h status check, T+24h satisfaction survey. Tone-matched to the issue severity.

Vercel Cron Twilio SMS Scheduled
Use cases

Three real tickets, resolved in seconds.

Emergency

Water leak, upper floor, 11pm

Classified as plumbing · emergency · landlord. Routed to the on-call vendor in zone. Tenant receives confirmation in 8 seconds with an arrival window.

Response time8s
Vendor dispatched< 30min
Human touches0
High priority

HVAC not cooling, family of 4

Classified as HVAC · high · landlord. Two top-rated vendors surfaced with slot availability. Manager approves with one click, tenant updated automatically.

Est. cost$380–$620
Vendors ranked3
Approval clicks1
Tenant-side

Clogged drain, non-urgent

Classified as plumbing · medium · tenant responsibility. Auto-response explains the assignment with lease clause reference. No vendor dispatched.

Auto-resolvedYes
Lease clause cited§4.2
Cost to landlord$0
Why real estate

The data picked the niche.

Before building, we scored 7,053 Upwork AI jobs across 22 industries. Real estate came out on top — 3.7× the demand of the next vertical.

7,053
Upwork AI jobs analyzed across 22 industries over a 60-day window.
3.7×
Real estate demand vs the next vertical. 48 jobs vs 13 in HVAC.
51.6
Opportunity score — real estate 51.6 vs HVAC 16.3. The gap is measurable.
Built with

Stack you already trust.

Next.js 15 TypeScript Claude Sonnet 4.6 Anthropic SDK Supabase pgvector Vercel Cron Twilio Tailwind CSS Tool use Row Level Security
FAQ

Questions, honestly answered.

Is this production-ready today? +
Agent 1 (Maintenance Estimator) is fully built on Claude Sonnet 4.6 with tool use and runs end-to-end. Agents 2 and 3 are in active development. A production pilot is planned for Q2 2026 with a mid-size property manager.
How much does it cost to run per ticket? +
Under $0.03 in Claude tokens per ticket on average, plus Twilio SMS fees (~$0.01 per outbound message). For a 200-unit portfolio averaging 40 tickets/month, infrastructure costs land under $15/month.
What about data privacy and tenant information? +
All tenant data is stored in a dedicated Supabase schema with Row Level Security enabled. No data is used for model training — Anthropic's zero data retention policy applies to API calls.
Can I customize the agents for my portfolio? +
Yes. The Estimator's cost ranges, the Router's skill taxonomy, and the Comms message templates are all configurable per-tenant. The vendor database is fully yours — seed it with your existing network.
Who built this? +
Richard Kennedy — AI Engineer and Agent Developer specializing in Claude-based systems. Based in Bandung, Indonesia. Building production multi-agent workflows for 50-500 unit property managers.

Let's talk property.

If you manage 50–500 units and want a look at what end-to-end maintenance automation costs, book a call.

Hire on Upwork richardbotsiokennedy@gmail.com