🤖 Erpy-Middleware: The Agentic ERP Interface

Erpy-Middleware is a high-performance agentic bridge providing a natural language interface to 180+ ERP and CRM systems.

🎯 Problem & Solution

Modern businesses are trapped in software silos—HubSpot for leads, QuickBooks for finance, NetSuite for ERP. Moving data between them requires complex manual entry or fragile, hard-coded integrations.

Erpy-Middleware solves this by inserting an intelligent reasoning layer between the user and their software. Using Gemini 2.5 Pro and Unified.to, it understands human intent and executes cross-platform actions autonomously, essentially turning any legacy CRM/ERP into a conversational partner.

---

🏗 Architecture & Data Flow

graph TD
    User([User]) -- "Natural Language" --> UI[Next.js Dashboard]
    UI -- "REST API" --> Backend[FastAPI Middleware]
    
    subgraph "AI Workspace"
        Backend --> Router{Gemini 2.5 Flash Router}
        Router -->|Analytics| Analyst[BI Analyst Agent]
        Router -->|Discovery| Discovery[Metadata Expert]
        Router -->|Automation| Automation[Action Agent]
    end
    
    subgraph "Execution Layer"
        Analyst --> DB[(Local SQLite Sync)]
        Discovery --> Unified[Unified.to API]
        Automation --> Unified
    end
    
    Unified <--> External[180+ ERP/CRM Systems]

Loading

---

🚀 Core Features & Workflows

1. Automated Lead & Contact Management

• Workflow: Use natural language to create, update, or retrieve contacts across multiple CRMs.
• Example: "Create a new contact in HubSpot for Jane Doe (jane@acme.com) and set her status to 'Qualified Lead'."
• Outcome: The Automation Agent maps the request to the HubSpot schema and executes the API call via Unified.to.

2. Intelligent Data Discovery & Mapping

• Workflow: Automatically analyze raw API metadata from any connected system to discover custom fields.
• Example: "Identify all custom fields in our NetSuite environment related to project costs."
• Outcome: The Discovery Expert analyzes the universal schema and provides a human-readable mapping strategy.

3. Native BI Chat (Local Analytics)

• Workflow: Chat directly with synced data (Customers, Products, Invoices).
• Example: "Who is our top-spending customer this month?" or "Summarize the last 5 unpaid invoices."
• Outcome: The Analyst Agent queries the local synchronized database to provide instant insights without extra API calls.

---

💬 Example User Prompts & Actions

User Prompt	Classified Intent	System Action
"Show me a summary of our total revenue."	ANALYTICS	Queries local invoices table and summarizes.
"What fields are available in my Salesforce connection?"	DISCOVERY	Fetches metadata from Unified.to and maps schema.
"Update the email for John Smith to js@new.com in HubSpot."	AUTOMATION	Formats payload and PATCHES the record via Unified.to.
"Create a new product 'Quantum Widget' with price 500."	AUTOMATION	Creates record in the connected Accounting ERP.

---

🛠 Tech Stack

• Frontend: Next.js 14, React, TailwindCSS/Vanilla CSS (UX-focused responsive design).
• Backend: FastAPI (Python), Pydantic-AI (Agent orchestration).
• AI Core: Google Gemini 2.5 Flash (Routing/BI) & Gemini 2.5 Pro (Reasoning/Discovery).
• Platform: Unified.to for universal data access.

---

⚙️ Environment & Configuration

Create a .env file in the backend/ directory with the following variables:

Variable	Description	Source
GEMINI_API_KEY	Google AI Studio Key	AI Studio
UNIFIED_API_KEY	Unified.to API Key	Unified.to Dashboard
DATABASE_URL	Local SQLite path	Default: sqlite:///./app/dummy.db

---

🛡 Security & Assumptions

• Data Privacy: All AI reasoning is performed using the user's provided API keys. Data remains in the local database or transit via Unified.to.
• Prototype State: This is currently an MVP/Prototype. The agents are highly capable but assume "Standard Objects" are mapped via Unified.to.
• Rate Limits: Performance is subject to Gemini API and Unified.to rate limits.

---

🗺 Roadmap

• Phase 1: Real-time Webhooks for instant sync between disparate ERPs.
• Phase 2: Multi-step "Workflows" (e.g., "If an invoice is paid in QuickBooks, email the contact in Salesforce").
• Phase 3: Support for "Custom Documents" and native file attachments (Gemini Multimodal).
• Phase 4: Enterprise SSO and Role-Based Access Control (RBAC).

---

🚦 Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+

Quickstart

1. Clone & Backend Setup

   git clone https://github.com/Jawknee-builds/Erpy-Middleware.git
   cd Erpy-Middleware/backend
   pip install -r requirements.txt
   python -m uvicorn app.main:app --reload

2. Frontend Setup

   cd ../frontend
   npm install
   npm run dev

---

Built with 🤖 and ❤️ by Jawknee-builds