Powered by CData Connect AI — A natural-language BI dashboard that unifies data across any combination of enterprise sources (CRM, ERP, Support, HRIS, Product Usage, Marketing) and lets business users query all of them in plain English.

BI Command Center sits on top of CData Connect AI. You connect your existing data sources through CData (Salesforce, HubSpot, QuickBooks, Zendesk, Snowflake, Google Sheets — anything CData supports), map them to entity categories in this app, then ask business questions in plain English.

The app automatically identifies which databases to query, writes the SQL, executes it across all sources in parallel, joins the results, and produces an executive-style summary.

Example questions you can ask:

• "Show me all enterprise accounts with open support tickets and outstanding invoices"
• "Which customers have high product usage but haven't renewed their contract?"
• "What's our revenue by region compared to support ticket volume?"
• "List employees who manage accounts with churn risk"

See ARCHITECTURE.md for the full system diagram, query execution flow, and security model.

flowchart TB
    subgraph BROWSER["🌐  Browser  (Client)"]
        direction LR
        UI["Login · Dashboard · Settings"]
        LS["🔒 LocalStorage — encrypted per user<br/>CData Connect AI credentials · LLM key · mappings · table selections"]
    end

    subgraph SERVER["⚙️  Next.js API Server"]
        direction LR
        QUERY["POST /api/query"]
        OTHERS["GET /api/sources<br/>GET /api/cdata/catalogs  ·  schemas"]
    end

    subgraph LLM_BOX["🤖  LLM Provider"]
        direction LR
        PLAN["Query Planner<br/>(schema → SQL plan)"]
        SYNTH["Result Synthesizer<br/>(rows → insights)"]
    end

    subgraph CDATA["☁️  CData Connect AI"]
        direction TB
        META["REST Metadata API<br/>catalogs / schemas / tables / columns"]
        MCP["MCP Endpoint — SQL execution"]
        subgraph SOURCES["Connected Data Sources"]
            direction LR
            CRM_N["🔵 CRM"]
            ERP_N["🟢 ERP"]
            SUP_N["🟡 Support"]
            HR_N["🩷 HRIS"]
            USG_N["🟣 Usage"]
            MKT_N["🟠 Marketing"]
        end
    end

    BROWSER -- "HTTPS" --> SERVER
    SERVER -- "① schema" --> META
    SERVER -- "② plan" --> PLAN
    SERVER -- "③ SQL" --> MCP
    SERVER -- "④ synthesize" --> SYNTH
    META --- SOURCES
    MCP --- SOURCES

    style BROWSER fill:#0f172a,stroke:#3b82f6,color:#dbeafe
    style UI fill:#1e3a5f,stroke:#3b82f6,color:#bfdbfe
    style LS fill:#1e293b,stroke:#475569,color:#94a3b8
    style SERVER fill:#1e1b4b,stroke:#6366f1,color:#e0e7ff
    style QUERY fill:#312e81,stroke:#6366f1,color:#c7d2fe
    style OTHERS fill:#312e81,stroke:#6366f1,color:#c7d2fe
    style LLM_BOX fill:#042f2e,stroke:#0d9488,color:#ccfbf1
    style PLAN fill:#0f3d3b,stroke:#14b8a6,color:#99f6e4
    style SYNTH fill:#0f3d3b,stroke:#14b8a6,color:#99f6e4
    style CDATA fill:#052e16,stroke:#16a34a,color:#dcfce7
    style META fill:#14532d,stroke:#22c55e,color:#bbf7d0
    style MCP fill:#14532d,stroke:#22c55e,color:#bbf7d0
    style SOURCES fill:#052e16,stroke:#15803d,color:#d1fae5
    style CRM_N fill:#1e3a8a,stroke:#3b82f6,color:#bfdbfe
    style ERP_N fill:#14532d,stroke:#22c55e,color:#bbf7d0
    style SUP_N fill:#78350f,stroke:#d97706,color:#fde68a
    style HR_N fill:#831843,stroke:#db2777,color:#fbcfe8
    style USG_N fill:#4c1d95,stroke:#7c3aed,color:#ddd6fe
    style MKT_N fill:#7c2d12,stroke:#ea580c,color:#fed7aa

The LLM never connects to your database directly. It only sees metadata (table/column names) during planning, and the final result rows during synthesis.

Ask a single question spanning multiple enterprise systems. The app queries all relevant sources in parallel and joins the results automatically.

Multiple users can share one browser. Each user's credentials, LLM key, connection mappings, and table selections live in separate localStorage namespaces keyed by their profile UUID. Switching users resets all state instantly — no page refresh required.

Zero hardcoded database connections. In Settings you browse your CData Connect AI catalogs, expand connections to see schemas, and assign each schema to one of six entity categories. Mappings are stored locally and sent to the server on every request as a base64-encoded header.

Pin specific tables within each source card. When tables are selected, the LLM only receives schema context for those tables — faster queries and more accurate SQL. Selecting nothing defaults to using all tables (up to 30 per connection).

CData Connect AI credentials and LLM API keys are encrypted with AES-GCM (key derived via PBKDF2 from the user's PAT) before being written to localStorage. They are never stored in plaintext. They are passed to the server only as HTTP request headers, per request.

Query results are displayed in a sortable data table and an auto-generated charts view (bar, line, or pie — chosen automatically based on data shape).

The Settings page shows live LLM token consumption: session tokens (since last login) and all-time totals, split by prompt vs. completion, with cost estimates per provider.

Save buttons in Settings are only active when unsaved changes exist. When configuration is up to date the button shows "✓ Saved" and is disabled — preventing accidental duplicate saves.

The app organises all data sources into six fixed categories:

One entity category can map to multiple CData connections/schemas. The same connection can be assigned to multiple entity categories.

• Node.js 18+
• A CData Connect AI account with at least one connection configured
• An API key for one LLM provider: Grok (xAI), Gemini (Google), OpenAI, Mistral, or DeepSeek

git clone https://github.com/dibyendud-cdata/bi-command-center.git
cd bi-command-center
npm install

cp .env.local.template .env.local

Only one value is required in .env.local:

CDATA_MCP_ENDPOINT=https://mcp.cloud.cdata.com/mcp
NEXT_PUBLIC_APP_NAME=BI Command Center
NODE_ENV=development

Important: Do not put your CData email, PAT, or LLM API key in .env.local. These are entered by each user through the app UI and stored encrypted in their browser. The server is stateless with respect to credentials.

npm run dev

Open http://localhost:3001.

On the login screen, enter your name, select a role, and enter your CData Username (Email) and Password (PAT). Click "Start".

Go to Settings → LLM Configuration. Select your provider (Grok, Gemini, OpenAI, Mistral, or DeepSeek), choose a model, and paste your API key. Click Save LLM Config.

Go to Settings → CData Connect AI Configuration. Click List Connections to fetch your available CData catalogs. Expand any connection to lazy-load its schemas. For each schema, click the entity buttons (CRM, ERP, Support, etc.) to assign it to one or more categories. Click Save Mappings.

Return to the dashboard. Your source cards in the sidebar will show the mapped connections and table counts. Type a question in the search bar or click a Quick Query Template.

User types a question in the search bar
        │
        ▼
POST /api/query
  Headers: X-CData-Email, X-CData-PAT, X-LLM-Provider, X-LLM-Key, X-LLM-Model,
           X-Connection-Mappings (base64), X-Table-Selections (base64)
        │
        ├── 1. buildSchemaContext()
        │       └── CData REST API: getTables() + getColumns() for each mapped entity
        │           • Filtered to selected tables if any pinned (otherwise up to 30 per connection)
        │           • Result cached for 5 minutes per user+mappings+selections combination
        │
        ├── 2. LLM planQuery(question, schemaContext)
        │       └── Returns JSON query plan:
        │           { sourcesNeeded, queries[{source, sql, canParallel}], joinStrategy }
        │
        ├── 3. executor.execute(plan, credentials, mappings)
        │       ├── Parallel queries → CData MCP endpoint (SQL execution)
        │       ├── Sequential queries → CData MCP endpoint
        │       └── In-memory LEFT/INNER JOIN of result sets
        │
        └── 4. LLM synthesizeResults(question, results)
                └── Returns { summary, insights[], followUpQuestions[] }

Response → { plan, results, synthesis, tokenUsage }

bi-command-center/
├── app/
│   ├── api/
│   │   ├── cdata/
│   │   │   ├── catalogs/route.ts     # List CData connections
│   │   │   ├── schemas/route.ts      # List schemas per connection
│   │   │   └── test/route.ts         # Connection health check
│   │   ├── sources/
│   │   │   ├── route.ts              # Source card status + table counts
│   │   │   └── [sourceId]/
│   │   │       ├── tables/route.ts           # Table list per entity
│   │   │       ├── tables/[tableName]/route.ts # Table data preview
│   │   │       └── preview/route.ts          # Quick preview
│   │   └── query/route.ts            # Main NL query endpoint
│   ├── settings/page.tsx             # Settings page
│   ├── page.tsx                      # Dashboard
│   ├── layout.tsx                    # Root layout + providers
│   └── globals.css
├── components/
│   ├── auth/
│   │   ├── LoginScreen.tsx
│   │   └── ProfileMenu.tsx
│   ├── query/
│   │   ├── SearchBar.tsx
│   │   ├── TemplateCards.tsx
│   │   ├── ResultsTable.tsx
│   │   ├── ResultsVisualizer.tsx
│   │   └── QueryPlan.tsx
│   ├── sources/
│   │   └── SourceList.tsx            # Source cards with table selection
│   ├── logo.tsx                      # SVG app logo
│   └── ui/                           # shadcn/ui components + toaster
├── contexts/
│   ├── auth-context.tsx              # Profile + credential management
│   ├── llm-config-context.tsx        # LLM settings + token tracking
│   └── mappings-context.tsx          # Connection mappings + table selections
├── lib/
│   ├── auth/storage.ts               # Encrypted profile localStorage
│   ├── cdata/
│   │   ├── client.ts                 # CData MCP query client
│   │   ├── metadata-client.ts        # CData REST metadata client
│   │   └── cache.ts                  # Query result cache
│   ├── llm/
│   │   └── config-storage.ts         # Encrypted LLM config localStorage
│   ├── mappings/
│   │   └── storage.ts                # Mappings + selections localStorage
│   ├── query/
│   │   └── executor.ts               # SQL execution + result join engine
│   └── llm-client.ts                 # LLM provider abstraction
├── types/
│   ├── sources.ts                    # DataSourceType, EntityMappings, etc.
│   └── queries.ts                    # QueryPlan, QueryResults, etc.
└── .env.local.template               # Environment variable reference

• Ensure your Username (Email) and Password (PAT) are correct in your profile
• Verify your connection appears in the CData Connect AI dashboard
• Go to Settings → expand a connection → assign schemas to entity categories

• Confirm your API key is saved (Settings → LLM Configuration → Save LLM Config)
• Check that the correct provider is selected for your key
• If queries time out, try a faster model (e.g. grok-3-fast, gemini-2.0-flash)

• Check which tables are selected in the source cards (empty = all tables)
• Try pinning only the relevant tables for your question
• The schema context is cached for 5 minutes — after remapping connections, wait or restart the server

Private — all rights reserved.