System Architecture

The "One-Man Army" tech stack powering the BackPocket Digital Twin.

BackPocket OS is built on a lightweight, high-performance architecture utilizing a localized "Brain" (FastAPI & SQLite) connected to powerful AI endpoints and communication channels.

Gmail API

Continuous polling for client emails, quotes, and supplier invoices.

BackPocket Brain

FastAPI (Python) + MCP
The central router utilizing the Model Context Protocol (MCP) to isolate AI tools into modular endpoints.

Gemini 2.5 Flash

Primary Agent for workflows. Now enhanced with PGVector RAG capabilities.

Moondream (Ollama)

Local AI Vision. Extracts data from uploaded site receipts and technical plans.

Flutter Mobile App

Native prototype app for Steve to manage the twin directly from his pocket.

WhatsApp API

Pushing 10-second approval loops to Steve on site via Whapi.

PostgreSQL + Vector

High-performance data storage and Vector RAG search capabilities.

Deployment Specifications

Local OS Agent: Local mini-PC setup running the `main.py` core to maintain data privacy.

Database: Upgraded to `PostgreSQL` for advanced Vector RAG search, synchronized to Google Sheets for client usability.

PWA & Mobile: Includes Progressive Web App (PWA) manifest and a dedicated Flutter native application.