SAPCopilotForge

An intelligent SAP ERP assistant with a built-in AI agent. The application combines a SAP Fiori interface with a conversational AI panel (SAP AI Copilot) that allows users to query SAP data using natural language purchase orders, materials, finances and invoices.

The AI agent analyzes user queries, classifies the relevant SAP module (MM, SD, FI), fetches the appropriate data and generates clear responses in Polish.

Screenshots

Dashboard with AI Copilot panel

AI agent conversation

Tech Stack

Layer	Technology
Frontend	SAPUI5 / OpenUI5 (sap_horizon theme, Fiori design)
Backend	Node.js + Express 5
AI Agent	Python + LangGraph + LangChain
LLM	Multi-provider: Anthropic, OpenAI, Google Gemini, Ollama
Agent API	FastAPI + Uvicorn

Frontend — SAPUI5 Fiori

• Native SAP Fiori interface built on OpenUI5 (CDN)
• sap_horizon theme — SAP's latest design system
• Views: Dashboard, Orders, Materials, Finance
• Built-in SAP AI Copilot panel — chat with the AI agent directly in the UI
• Natural language navigation from the chat (e.g. "show orders")
• Edit & resend previous messages in the chat
• Full i18n support (Polish)
• Components: tnt:ToolPage, SplitApp, GenericTile, IconTabBar, Table

Backend — Node.js Express

• Express 5 server serving static SAPUI5 files
• API proxy forwarding requests to the Python agent
• CORS support and global error handling
• Endpoints: /api/chat, /api/agent/query, /api/agent/create, /api/health
• Automatic fallback to mock data when the agent is offline

AI Agent — Python LangGraph

• LangGraph state graph with stages: intent classification → SAP data retrieval → response generation
• Multi-provider LLM — switch between providers with a single env variable
• FastAPI server with REST endpoints
• SAP module support: MM (purchasing/materials), SD (sales), FI (finance)
• Mock data layer simulating SAP system responses

Supported LLM Providers

Provider	LLM_PROVIDER	API Key Variable	Default Model	Install
Anthropic	anthropic	ANTHROPIC_API_KEY	claude-sonnet-4-20250514	included
OpenAI	openai	OPENAI_API_KEY	gpt-4o	included
Google Gemini	google	GOOGLE_API_KEY	gemini-2.0-flash	pip install langchain-google-genai
Ollama (local)	ollama	— (free)	llama3.1	pip install langchain-ollama

If LLM_PROVIDER is not set, the agent auto-detects the provider based on which API key is available.

You can override the model with LLM_MODEL in .env.local.

Requirements

• Node.js >= 18
• Python >= 3.10
• LLM API key — at least one of: Anthropic, OpenAI, Google, or local Ollama

Installation

1. Node.js dependencies

npm install

2. Python dependencies (AI agent)

npm run agent:install

3. LLM configuration

Copy the example and edit .env.local:

cp .env.example .env.local

Anthropic (default):

ANTHROPIC_API_KEY=sk-ant-YOUR_KEY

OpenAI:

LLM_PROVIDER=openai
OPENAI_API_KEY=sk-YOUR_KEY

Google Gemini:

LLM_PROVIDER=google
GOOGLE_API_KEY=YOUR_KEY

Ollama (free, local):

LLM_PROVIDER=ollama
LLM_MODEL=llama3.1

Running

You need two terminals running simultaneously:

Terminal 1 — Node.js server (port 3000)

npm run dev

Terminal 2 — Python AI agent (port 8000)

npm run agent:start

The application is available at: http://localhost:3000

Commands

Command	Description
npm run dev	Node.js server with auto-reload
npm run start	Node.js server (production)
npm run agent:start	Python AI agent (LangGraph)
npm run agent:install	Install Python dependencies

Architecture

graph TB
    subgraph Browser["Browser"]
        UI["SAPUI5 Fiori App<br/><i>sap_horizon theme</i>"]
        subgraph Views["Views"]
            V1["Dashboard"]
            V2["Orders"]
            V3["Materials"]
            V4["Finance"]
        end
        Chat["SAP AI Copilot<br/><i>Chat Panel</i>"]
        UI --- Views
        UI --- Chat
    end

    subgraph NodeServer["Node.js Express :3000"]
        Static["Static Files<br/><i>webapp/</i>"]
        API_Chat["POST /api/chat"]
        API_Query["POST /api/agent/query"]
        API_Create["POST /api/agent/create"]
        API_Health["GET /api/health"]
        Mock["Mock Data<br/><i>fallback offline</i>"]
    end

    subgraph PythonAgent["Python Agent :8000"]
        FastAPI["FastAPI Server"]
        subgraph LangGraph["LangGraph StateGraph"]
            Classify["Classify Intent<br/><i>MM · SD · FI</i>"]
            Fetch["Fetch SAP Data<br/><i>mock layer</i>"]
            Generate["Generate Response"]
            Create["Create Document"]
            Classify --> Fetch
            Fetch -->|query/analyze| Generate
            Fetch -->|create| Create
        end
        FastAPI --- LangGraph
    end

    subgraph LLM["LLM Provider"]
        Anthropic["Anthropic<br/><i>Claude</i>"]
        OpenAI["OpenAI<br/><i>GPT-4o</i>"]
        Google["Google<br/><i>Gemini</i>"]
        Ollama["Ollama<br/><i>local</i>"]
    end

    Browser -->|"HTTP"| NodeServer
    NodeServer -->|"static files"| Browser
    API_Chat -->|"proxy"| FastAPI
    API_Query -->|"proxy"| FastAPI
    API_Create -->|"proxy"| FastAPI
    Generate -->|"API call"| LLM
    Create -->|"API call"| LLM

    style Browser fill:#e8f4fd,stroke:#1a73e8,color:#000
    style NodeServer fill:#e6f4ea,stroke:#34a853,color:#000
    style PythonAgent fill:#fef7e0,stroke:#f9ab00,color:#000
    style LLM fill:#fce8e6,stroke:#ea4335,color:#000
    style LangGraph fill:#fff8e1,stroke:#f9ab00,color:#000
    style Views fill:#e8f0fe,stroke:#4285f4,color:#000

Loading

Request Flow

sequenceDiagram
    actor User
    participant UI as SAPUI5 Chat
    participant Node as Express :3000
    participant Agent as FastAPI :8000
    participant Graph as LangGraph
    participant LLM as LLM Provider

    User->>UI: types message
    UI->>Node: POST /api/chat
    Node->>Agent: proxy POST /chat
    Agent->>Graph: run_query()
    Graph->>Graph: classify_intent()
    Graph->>Graph: fetch_sap_data()
    Graph->>LLM: generate_response()
    LLM-->>Graph: AI response
    Graph-->>Agent: result
    Agent-->>Node: JSON response
    Node-->>UI: JSON response
    UI-->>User: display message

    Note over Node,Agent: If agent offline,<br/>Node returns mock data

Loading

Troubleshooting

"Cannot connect to AI agent"

• Make sure the Python agent is running (npm run agent:start)
• Check that your API key is set in .env.local

Agent won't start

• Install dependencies: npm run agent:install
• Check Python version: python --version (requires >= 3.10)

Wrong LLM provider

• Check active provider: curl http://localhost:8000/health
• Set LLM_PROVIDER explicitly in .env.local

Application doesn't load in browser

• Make sure the Node.js server is running (npm run dev)
• Open http://localhost:3000