Real Estate Voice AI: 10 Specialist Agents (CallSphere) vs Vapi Single-Agent

TL;DR

CallSphere Real Estate (live at realestate.callsphere.tech) runs 10 specialist agents orchestrated by a triage agent named Aria, built on the OpenAI Agents SDK with hierarchical handoffs. Vapi.ai's "Squads" feature can route between voice agents, but the multi-agent orchestration, the 30+ tools across the agents, the shared property/suburb/financial data layer, and the tenant lifecycle modeling are all yours to build. To replicate CallSphere's setup on Vapi, expect 10x the engineering time of a single-agent build. This post enumerates every agent, what it does, the tools it owns, and how the handoff graph keeps a real estate conversation coherent.

Why Real Estate Voice Needs Multiple Agents

A single inbound call to a brokerage can shift through five intents in three minutes:

• "I want to look at 24 Maple Street." (Property search)
• "What are the schools like there?" (Suburb intelligence)
• "What would my mortgage be at $620k?" (Mortgage)
• "Is this a good investment if I rent it?" (Investment)
• "Can I see it on Saturday?" (Viewing scheduler)

Cramming all of that into one prompt creates a confused, slow agent. Splitting into specialists gives each one a tighter prompt, a smaller toolset, and faster latency. The triage agent owns the conversation; specialists own the answers.

Vapi's Multi-Agent Story

Vapi supports "Squads" — multiple voice assistants on a single call, with handoffs. It is a real feature. What is missing relative to a real estate vertical:

• No domain agents. You design and prompt all 10 specialists yourself.
• No shared property graph. Each agent's tools point at databases you stand up.
• No vision tool. Vapi voice agents are voice-only.
• No suburb dataset. Schools, demographics, commute, forecasts — you source and ingest each.
• No tenant lifecycle model. Leases, rent ledger, maintenance — you build the schema and the tools.
• No emergency override agent. You design the escalation path.

Squads gives you the orchestration primitive. CallSphere gives you the orchestration plus the 10 agents, plus the data, plus the tools, plus the tenant lifecycle.

CallSphere's 10 Agents

1. Triage (Aria) — orchestrator. Listens for intent, hands off to a specialist, hands the conversation back when done.
2. Property Search — semantic + filter + vision. Listing match, photo analysis, "find me a 3-bed under $750k near light rail."
3. Suburb Intelligence — neighborhood profiles, schools, demographics, commute times, market forecasts.
4. Mortgage Calculator — affordability, repayments, current bank rates, loan term scenarios.
5. Investment Calculator — gross yield, net yield, cash flow, vacancy assumptions, rental comps.
6. Price Watch — value scoring, deal alerts, listings priced below comp average.
7. Viewing Scheduler — open homes, private viewings, calendar invites.
8. Agent Matcher — find a buyer/listing agent, reviews, contact info.
9. Maintenance — for tenants: create, track, escalate maintenance requests.
10. Payment — for tenants: rent balance, ledger, accept payments.
11. Emergency — separate agent that activates instantly for tenant emergencies (gas leak, flood, lockout). Treated as an override; bypasses triage.

Comparison Table

Capability	Vapi (Squads)	CallSphere Real Estate
Triage / orchestrator agent	DIY	Aria built-in
Property search agent	DIY	Built-in (semantic + filter + vision)
Suburb intelligence agent	DIY	Built-in
Mortgage calculator agent	DIY	Built-in
Investment calculator agent	DIY	Built-in
Price watch agent	DIY	Built-in
Viewing scheduler agent	DIY	Built-in
Agent matcher	DIY	Built-in
Maintenance agent (tenant)	DIY	Built-in
Payment agent (tenant)	DIY	Built-in
Emergency override agent	DIY	Built-in
Shared property graph	DIY	Built-in
30+ tools across agents	DIY	Built-in
Vision (photo analysis)	Not native	Built-in
Estimated buildout time	12+ months	Live

Agent Hierarchy Diagram

flowchart TD
    Caller[Caller] --> Aria[Triage: Aria]
    Aria --> PS[Property Search]
    Aria --> SI[Suburb Intelligence]
    Aria --> MC[Mortgage Calculator]
    Aria --> IC[Investment Calculator]
    Aria --> PW[Price Watch]
    Aria --> VS[Viewing Scheduler]
    Aria --> AM[Agent Matcher]
    Aria --> M[Maintenance]
    Aria --> P[Payment]
    Caller -. emergency keyword .-> EM[Emergency Agent]
    EM --> Dispatch[On-Call Dispatch]
    PS --> DB[(properties + listings + photos)]
    SI --> DB2[(suburb_profiles + school_zones)]
    MC --> RATES[(bank_rates)]
    IC --> RENTS[(rental_comps)]
    PW --> SCORE[(value_scores)]
    VS --> CAL[(viewing_calendar)]
    AM --> AGENTS[(agent_directory)]
    M --> MR[(maintenance_requests)]
    P --> RL[(rent_ledger)]
    PS -. handoff back .-> Aria
    SI -. handoff back .-> Aria

Worked Example: First-Home Buyer Discovery Call

Caller: "Hi, my partner and I are looking at 3-bedrooms under $700k somewhere with a good elementary school."

The call traverses 5 of the 10 agents in 6 minutes.

1. Aria classifies intent → property search + suburb + affordability bundle.
2. Property Search runs filter + semantic match → returns 14 candidates.
3. Suburb Intelligence layers school ratings on each candidate's suburb → narrows to 6.
4. Aria hands back to caller: "I have 6 that match. Want to talk affordability before viewing?"
5. Mortgage Calculator runs scenarios at 6.2% and 6.5% across $620k and $680k purchase prices, factoring in caller's stated income.
6. Caller picks two listings. Viewing Scheduler books Saturday open homes.
7. Post-call, call_log_analytics runs (sentiment +0.7, intent=buyer-discovery, lead=84/100, satisfaction=5).
8. The brokerage's CRM gets a hot lead with full transcript and the 6 listing IDs.

On Vapi, that workflow is 5 separate agent prompts, 5 tool sets you build, a shared property database, and a handoff state machine you debug. Live in 9-12 months. On CallSphere, live the day you go live.

Migration / Decision Section

The decision tree:

• You are a brokerage, property manager, or proptech buyer. CallSphere — the 10 agents and 30+ tools cover the workflow.
• You are a developer building a real estate proptech product whose differentiation is the agent design itself. Vapi — you want full control of every prompt and turn.
• You are an enterprise franchise with a custom CRM and bespoke regional data. Hybrid — CallSphere as the front-line voice, custom Vapi agents for niche overrides.

Most brokerages, property management firms, and proptech buyers we onboard pick CallSphere because the build math doesn't work otherwise.

FAQ

Are the 10 agents customizable?

Yes. Prompt phrasing, tool gating, and persona voice/personality are configurable per brokerage. You can also disable agents you don't need (e.g., a sales-only brokerage may turn off Maintenance and Payment).

How does emergency routing work?

The Emergency agent is keyword- and intent-listening throughout the call. If a tenant says "gas leak," "fire," "flood," "lockout," etc., the conversation immediately reroutes to Emergency, which collects address + nature, dispatches per the property's runbook, and notifies on-call staff.

What model powers the agents?

GPT-4o-realtime-preview for voice; GPT-4o-mini for analytics; the OpenAI Agents SDK orchestrates handoffs. The 6-container pod includes a Go gateway, an AI worker, a voice server, NATS, Redis, and the Next.js frontend.

Can I bring my own MLS feed?

Yes. CallSphere supports common MLS sync patterns (RESO Web API, IDX feeds) and refreshes the property graph on a configurable cadence. Custom feeds are supported on enterprise plans.

What about non-US/AU markets?

The architecture is market-agnostic; the suburb dataset and bank rate sources are configured per region at deploy time. Active deployments span US and AU; other markets onboard via custom data ingestion.

How is the handoff different from Vapi Squads?

CallSphere's handoffs are typed via the OpenAI Agents SDK — each agent declares what it accepts and what it returns. The triage agent reasons about handoff destinations using the SDK's runtime, not via prompt-stitching. That produces measurably more reliable transitions in long calls.

Hear all 10 agents in a real conversation at /demo or read the deeper stack at /industries/real-estate.