Teleta Vox

The AI-powered CRM built for real estate professionals.

Voice-first client management, automated follow-ups, receipt scanning, and a concierge-grade interface that finally makes brokers want to open their CRM.

---

What it does

Teleta Vox is an end-to-end operating system for independent agents and small brokerages. Instead of making a user navigate a twelve-tab CRM, they hold down a button and talk. The agent understands intent, routes the request through the correct tool, and writes back to MongoDB.

A working conversation might sound like:

"Log a new buyer lead — Sarah Chen, three-bed in Capitol Hill, budget seven-fifty, saw her at the Sunday open house. Remind me to follow up Tuesday morning."

That single utterance creates a client record, attaches a note, schedules a task, and files the source attribution — no typing, no forms.

---

Core features

Area	What it does
Voice assistant	ElevenLabs streaming STT + TTS over WebSocket, wired to a Pydantic AI agent. Drives every CRUD operation in the app.
Client CRM	Pipelines, notes, tagged sources, compliance disclosure tracking, import from CSV.
Tasks & scheduled follow-ups	APScheduler worker fires reminder emails and in-app nudges on the agent's behalf.
Inbox & calendar	Gmail + Google Calendar API integration — read, triage, reply, and create events from voice.
Receipt scanning & tax ledger	Snap a photo → vision model extracts line items → stored against a deductible category for year-end export.
Mileage log	Tap-to-track driving sessions with route geometry, auto-calculated IRS-rate mileage.
Google Drive storage	drive.file scope — the user owns the files, we never see anything outside what we created.
Compliance chat (RAG)	Qdrant vector store seeded with state-level real-estate law; agent retrieves citations before answering legal questions.
Billing	Stripe Checkout + webhook-driven plan upgrades. Solo / Pro, monthly or yearly.

---

Who it's for

Teleta Vox is built for the solo real-estate agent and the two-to-ten person boutique brokerage — the operators who live out of their phone, drive between showings, and have neither the time nor the stomach for a Salesforce implementation.

The target user is someone who:

• Runs their book of business on sticky notes, screenshots, and the iPhone Notes app.
• Loses deductions every April because receipts vanished into a glovebox.
• Doesn't want to become a CRM administrator — they want to sell houses.
• Would happily talk to software if the software actually understood them.

We deliberately do not target enterprise brokerages with existing tech stacks. Our north-star customer is the producer who has outgrown a spreadsheet but refuses to be trained on Follow Up Boss.

---

Architecture at a glance

          ┌──────────────────────┐      ┌──────────────────────┐
          │   Next.js 16 (App)   │◄────►│   FastAPI (Python)   │
          │   Auth.js · Tailwind │ SSE  │   Pydantic AI agent  │
          └──────────┬───────────┘      └──────────┬───────────┘
                     │                              │
          ┌──────────▼──────────────────────────────▼──────────┐
          │              MongoDB Atlas (shared)                │
          │   users · sessions · clients · tasks · receipts    │
          └────────────────────────────────────────────────────┘
                     │                              │
          ┌──────────▼──────────┐       ┌──────────▼──────────┐
          │   Qdrant (vector)   │       │  ElevenLabs · Gemini│
          │    compliance RAG   │       │  Stripe · Resend    │
          └─────────────────────┘       └─────────────────────┘

Frontend — Next.js 16 App Router, route-grouped dashboard, Auth.js v5 (Google / LinkedIn / Resend magic link), Tailwind + Framer Motion for the editorial motion language.

Backend — FastAPI with a single Pydantic AI agent that owns every tool function. Motor for async Mongo, APScheduler for the email worker, Qdrant for vector retrieval.

AI layer — OpenRouter routes to Gemini 2.5 Flash. Streaming SSE back to the frontend. A VoxDeps context object carries the user ID, timezone, plan, and session summary on every tool invocation, so the agent never confuses whose data it's touching.

Auth model — no separate backend auth layer. FastAPI reads the Auth.js session cookie and looks it up in Mongo's sessions collection. One source of truth, zero token-sync bugs.

---

Infrastructure

We run the whole thing on a DigitalOcean VPS managed by Dokploy — a self-hosted Heroku/Vercel alternative. Dokploy points at our Docker Compose file, watches the Git repo, and handles zero-downtime redeploys, Let's Encrypt certs, and env-var injection.

• Frontend → teletavox.com
• Backend → api.teletavox.com
• Qdrant → named Docker volume, persisted across deploys
• Database → MongoDB Atlas (three environments: local_testing / development / production, switched by DB name in the URI)

This setup keeps our monthly infra bill under the cost of a single Vercel Pro seat, while giving us full control over the Python backend and the vector store.

---

About the company

Teleta Vox is a two-founder, three-person company.

Ethan Breitkreutz — Co-founder

Dave Alton — Co-founder

Advaith Madhus — Founding Engineer

_{↑ replace with a team photo}

Why we're building this

Real estate software is stuck in 2008. The incumbents sell by the seat to brokerages, which means the product serves the broker-owner, not the agent. We're flipping that: Teleta Vox is bought by the agent, lives on their phone, and treats the CRM as a byproduct of them doing their job — not another job on top of it.

---

teletavox.com  

_{This repository is a showcase. The production source lives in a private monorepo.}