Clara AI — operator, not chatbot

Clara is your AI property operator — maintenance, leasing, and tenant conversations handled around the clock.

One persistent voice across SMS, email, and voice. Grounded in your live data, built for Fair Housing compliance.

Inside the platform

One operator, any channel

Voice, SMS, email — the action is the same.

Clara doesn't care which channel the trigger arrives on. Same dispatch, same work order, same tenant confirmation.

TriggerSMS
AC is leaking onto the floor in unit 4B. Can someone come?
2:14 AM

Who Clara is, in three paragraphs

Warm, helpful, clear. Never bubbly.

01Persistent persona, not a chat window.

The same voice answers the 11 PM leak, the Saturday prospect, and the Tuesday rent-increase call. She never breaks character — not for an angry tenant, not for a clever prompt. Every conversation opens with “Clara here, the AI assistant for [your property].” Honesty is the architecture, not the tone.

02Warm, without being cute.

Clara talks the way the friendly neighbor who knows the building talks — contractions, short sentences, no corporate filler. No mascot, no exclamation-point energy. The target register is the leasing director who calls back at 7 AM with the answer already formed: direct, names the unit number, quotes the rate. That works for a stressed tenant at midnight as much as a prospect on a Saturday.

03Tool-grounded by construction.

Clara only quotes facts she pulled live — pricing from your rent roll, availability from the live calendar, a citation from the appliance manual for that specific unit. If the data is silent, she says so and offers a handoff. The system forbids inference where the data is silent. That single rule is why operators trust her with the keys to the inbox.

Watch one thread, end to end

A real maintenance scenario, with the tool calls and safety checks alongside.

The same tool chain Clara actually runs in production — labelled so you can show your team what 'AI operator' means in concrete terms.

Illustrative thread. Sarah Martinez, Unit 312; Mile High Plumbing rates and Moen SKU from our seed data.

  1. Turn 111:42 PMSarah

    Inbound, before any AI sees it

    The inbound text arrives. STOP / START / HELP keywords are intercepted before the model sees them and answered by deterministic code — not the AI. The consent record from /sms-consent (IP, user agent, ISO-8601 timestamp) is checked before the message is forwarded to Clara.

    Safety layer TCPA interception + consent ledger

  2. Turn 211:42 PMClara

    First turn always identifies as AI

    The caller-identification lookup matched Sarah Martinez in Unit 312 from the inbound phone number. Clara opens with her name, her role, and the property she manages — never role-plays as a human leasing agent. Honesty is an architectural rule, not a tone preference.

  3. Turn 311:43 PMSarah

    Mode is locked on turn one

    This thread was classified MAINTENANCE on Sarah’s first inbound. It cannot drift into a leasing tour or a renewal pitch — even if Sarah pivots topics, the leasing tools are not loaded into Clara’s context for this conversation. One locked mode per thread.

    Safety layer Two locked modes

  4. Turn 411:43 PMClara

    Quotes from real data, not memory

    The unit appliance inventory returned the Moen SKU and install date. The manual search is running against the live manufacturer PDF for the shut-off step. If either tool returned no result, Clara would say so and offer a handoff — guessing is disallowed at the prompt layer.

  5. Turn 511:44 PMClara

    Self-serve first, dispatch second

    The leaking-faucet playbook returned the manual-shutoff steps. The vendor lookup returned Carlos Reyes (Mile High Plumbing, $125/hr after-hours, on-call). Self-serve is always attempted before a paid dispatch — that’s what drives the 25–40% reduction in paid work orders Clara books for operators.

  6. Turn 612:18 AMClara

    The audit trail is the product

    The work order was persisted with the full tool-call chain — every call, every result, every timestamp. The vendor ETA wrote back to the property manager’s calendar. The thread itself — every Clara turn, every tool result, the photo, the manual citation — lands in AppFolio under Sarah’ ledger before the PM’s 7 AM coffee.

    Safety layer Full audit trail

Every Clara reply also passes a post-generation Fair Housing classifier before it reaches the tenant’s phone. The classifier blocks unsafe outputs server-side; the model never gets the final word. Read the spec below.

The other mode

Leasing threads work the same way — same architecture, different register.

Clara's mode is locked on the first message of every conversation. A maintenance thread cannot drift into a leasing pitch; a leasing thread cannot open a work order.

What Clara does on a prospect thread.

  1. 01
    Look up the property a prospect is asking about"Yes, Parkside Residences is the 50-unit building at 2200 S Yale. 1BR and 2BR, pet-friendly, rooftop."
  2. 02
    Surface real units, real pricing, real concessions"Unit 312 is a 2BR at $2,395 with a $500 concession through May 30 — not rounded, not a range."
  3. 03
    Walk through amenities and the neighborhood"Rooftop with grills, in-unit laundry, covered parking at $85/mo. You're four blocks from the light-rail stop."
  4. 04
    Capture and update the prospect record"Got your move-in date as July 15 and bedroom count as 2 — I'll send the lead summary to our leasing team tonight."
  5. 05
    Check the live tour calendar"Saturday at 10:30 AM or 2:00 PM is open on the leasing director's calendar. Either work?"
  6. 06
    Schedule, reschedule, or cancel a tour"Confirmed: Saturday, 10:30 AM at Parkside Residences. I'll text a reminder Friday evening."

Put Clara on your worst inbox night.

The thread demo above is illustrative. The next one will be real, in your portfolio, with your tenants. White-glove onboarding, direct input on the roadmap.

Read the platform overview