Super Realty

The always-on voice receptionist for solo real estate agents. It answers every call in your name, qualifies the buyer, books the showing, and never forgets a caller.

Every missed call is a missed commission: buyers dial the agent who answers first, and most calls come in after hours. Super Realty picks up every one, 24/7, in your name and voice, and remembers every buyer across calls.

Run it · How it's built · Work with me

What it does

• Answers in your name and voice. A voice agent picks up every call, discloses recording, and qualifies budget, timeline, financing, and area. It only ever describes homes you have connected.
• A team of specialists, not one bot. Each call is handled by a Concierge that greets and routes, a Property agent that searches and shows homes, and a Scheduling agent that books, handing off in real time. The Super Agents dashboard shows the handoff live.
• Onboards from one URL. Paste your website; it extracts your listings and infers your persona (name, agency, area, tone) for you to review.
• Remembers every buyer, across calls. Buyers and homes live in a Cognee knowledge graph. A returning caller is greeted by name and picks up where they left off.
• Books real showings. It checks your Cal.com calendar and books, with an idempotency key so a retry never double-books.
• Hands you the lead instantly. When the call ends it texts the buyer and outcome to your phone, so you follow up while the lead is warm.

Run it

There is no hosted demo. Run the whole stack yourself in one command:

git clone https://github.com/mahimairaja/superrealty.git
cd superrealty
make up

make up bootstraps .env, builds and starts everything (Postgres + pgvector, Neo4j, a bundled LiveKit server, backend, agent, frontend), applies migrations, and seeds a demo. Frontend at http://localhost:5173, API docs at http://localhost:8000/docs.

For voice, add OPENAI_API_KEY and DEEPGRAM_API_KEY to .env. To talk to the agent in your terminal, no LiveKit needed:

docker compose run --rm agent uv run python main.py console

Prefer plain Docker? docker compose up works too. See .env.example for every setting.

How it's built

One voice agent runs the whole call: identity, listing search, lead capture, booking, and forget are tools on a single agent behind one tenant-scoped API, so the same rules apply whether a buyer speaks or types. Memory is the system of record: listings and buyers live in Cognee's graph and vectors, which is what turns a returning caller into a known buyer instead of a fresh transcript.

Stack: LiveKit voice, Deepgram + OpenAI, Cognee memory (Neo4j + pgvector), FastAPI, React 19, Cal.com + Telnyx, multi-tenant, MIT.

Work with me

Built by Mahimai. One developer, end to end: the voice agent, the memory graph, the telephony, the multi-tenant backend, and the UI. I ship production voice agents on an open-source stack. Want one for your business?

• Book a call → mahimai.ca/contact
• Email → contact@mahimai.ca

License

MIT. You own what you run.