Homzy — AI Property Advisor (Phase 1: the Brain)

A bilingual (Arabic / English) AI "broker" that understands a client's needs and pitches matching properties from your real listings — never invented prices.

This is Phase 1 (the brain + chat screen) plus Phase 2 (a built-in admin panel to add/edit/delete your listings). Phase 3 = the Android/iOS mobile app.

Continuing the project, or picking it up on another device? Read docs/PROJECT_PLAN.md for the full plan, decisions, brand system, and the complete roadmap — and CLAUDE.md for the AI-assistant operating guide. The repo is self-contained: clone it and go.

Quick start (easiest)

Double-click run.bat. It sets up Python, installs everything, and opens the chat at http://127.0.0.1:8000.

The first time, it runs in Preview mode: replies are templated but already use your real listings and prices, so you can see the matching working. To turn on full, natural AI conversation, pick one of the two free engines below.

Deploy online (Vercel)

The repo is Vercel-ready (vercel.json + api/index.py wrap the FastAPI app):

Push this repo to GitHub (already done).
On https://vercel.com → Add New → Project → Import this repo. Framework preset: Other. Deploy. You get a public URL serving the web chat + API.
AI engine in the cloud: Ollama can't run on Vercel, so the hosted app defaults to template/preview mode. For real AI replies, add Project → Settings → Environment Variables: LLM_PROVIDER=gemini and GEMINI_API_KEY=<free key from https://aistudio.google.com/apikey>, then redeploy.

Caveats on Vercel: the filesystem is read-only, so admin-panel edits to data/listings.json won't persist, and in-memory chat sessions aren't durable across serverless instances. For persistent edits / durable sessions, host the container on Render / Railway / Fly.io, or back the listings with a database.

Turn on the AI brain (pick one — both free)

Option A — Ollama (local, free, no key, private) — recommended

Install Ollama for Windows: https://ollama.com/download
Open a terminal and pull a bilingual model (one-time, a few GB):
```
ollama pull qwen2.5:7b
```
(On a lighter PC use qwen2.5:3b and set OLLAMA_MODEL=qwen2.5:3b in .env.)
Make sure Ollama is running, then start Homzy with run.bat. The status pill turns green ("AI: ollama").

Option B — Google Gemini (free cloud tier, needs a free key)

Get a free key: https://aistudio.google.com/apikey

Copy .env.example to .env and set:

LLM_PROVIDER=gemini
GEMINI_API_KEY=your_key_here

Install the client: pip install google-generativeai
Start Homzy with run.bat.

No paid API is used anywhere. Your Anthropic/Claude key is not required.

Editing the broker and the listings

What	Where
Broker name / brand	`.env` (`BROKER_NAME`, `BRAND_NAME`)
Personality & sales style	`backend/persona.py` → `BROKER_TEMPLATE`
The properties & prices	`data/listings.json`
How many options to show	`.env` (`MAX_RESULTS`)

After editing data/listings.json, just restart run.bat.

Or use the admin panel (no JSON editing)

Open http://127.0.0.1:8000/admin (or click the ⚙︎ in the chat header) to add, edit, and delete properties in a form — bilingual fields included. Changes save straight to data/listings.json and the broker picks them up immediately, no restart needed. To password-protect the panel, set ADMIN_TOKEN in .env.

How "never invent a price" is guaranteed

The AI never makes up properties. Each turn, the code reads the client's needs, searches data/listings.json itself, and hands the model only the real matching rows. The model may only present what it's given — prices come from your file, not from the model.

Project layout

Homzy/
├─ backend/
│  ├─ app.py        FastAPI server + chat API
│  ├─ broker.py     per-turn pipeline (language, needs, match, reply)
│  ├─ listings.py   load + search the inventory (source of truth for prices)
│  ├─ persona.py    the broker's character + prompts + preview templates
│  ├─ llm.py        free engines: Ollama / Gemini / mock
│  └─ config.py     settings from .env
├─ data/listings.json   your properties (sample data — replace with real)
├─ frontend/index.html  the chat screen
├─ frontend/admin.html  the listings admin panel (/admin)
├─ run.bat              one-click setup + launch
└─ requirements.txt

The listings in data/listings.json are sample placeholders — replace them with your real units (or we'll wire up file upload in Phase 2).

Homzy — AI Property Advisor