Realtor AI
Realtor AI is a premium AI personal assistant for real estate agents. It recommends follow-ups, drafts messages from iMessage and Gmail context, and routes every send through explicit realtor approval via Telegram.
- Timeline
- 20 weeks from prototype agent to approval-first production workflow
- Role
- AI architecture, macOS agent development, Telegram control surface, and integration design
The challenge
Top-producing real estate agents manage dozens of active client relationships across iMessage, email, and lead platforms — but follow-up is the first thing that slips when showings stack up. Generic CRM reminders feel noisy, and AI tools that auto-send messages are a non-starter in a trust-driven industry. Agents needed an assistant that understood their conversation history, drafted context-aware messages, and waited for explicit approval before anything reached a client.
Our approach
We built Realtor AI as an approval-first loop: recommend, draft, preview, rewrite, approve. A dedicated Mac mini bridges iMessage by reading local thread history and sending through Messages.app — the same channels clients already use — while Gmail OAuth monitors trusted lead platforms for new inquiries. The Telegram bot serves as the control surface for daily briefs, client summaries, and drafted follow-ups, keeping the agent in the loop without opening another dashboard. Voice-note transcription and persistent client notes let agents capture context on the go.
The solution
Realtor AI combines a local macOS assistant agent with a Telegram interface, OpenAI-powered conversation analysis, and drafting tuned for real estate tone. The system ingests iMessage threads and Gmail lead notifications, maintains a client store with notes and engagement history, and runs a re-engagement cadence engine that flags relationships going quiet. Every outbound message passes through a human-reviewed queue — the agent sees the draft, can rewrite it, and taps approve before anything sends. Architecture documentation maps the production path: a cloud control plane, web dashboard, and per-realtor hosted Mac mini for scale.
Outcomes
Agents using Realtor AI completed more follow-ups within the window where clients are most responsive, while spending less time writing routine check-ins. The approval-first model built trust — agents felt augmented, not replaced — and daily briefs caught stale leads that would previously have slipped through during busy weeks. The architecture roadmap gives a clear path from single-agent prototype to a hosted platform serving entire brokerages.
Key outcomes
- More follow-ups completed within the optimal response window
- Less time spent drafting routine client messages
- Approval-first workflow ensured agents retained full control over every outbound communication
- Daily briefs surfaced stale leads and upcoming tasks before opportunities went cold