AI Features Overview

Drizzle Cube provides two approaches for AI-powered analytics:

Built-in MCP Endpoints - Zero-config AI readiness using server-side NLU (no LLM required)
Custom AI Endpoints - Full LLM-powered query generation (you implement with your preferred provider)

Built-in MCP Server

All framework adapters include a built-in MCP server at /mcp that lets AI agents discover and query your data:

Tool	Purpose
`drizzle_cube_discover`	Find relevant cubes based on topic or intent
`drizzle_cube_validate`	Validate queries and get auto-corrections
`drizzle_cube_load`	Execute queries and return results

Connect Claude, ChatGPT, n8n, or any MCP-compatible client to your semantic layer.

Learn more about the MCP Server →

Custom AI Endpoints

For more sophisticated natural language understanding, you can implement custom AI endpoints using an LLM provider like Google Gemini.

Note: Custom AI endpoints (/api/ai/*) are not included in the drizzle-cube package. You must implement them yourself. We provide prompt templates, type definitions, and a complete reference implementation.

What Drizzle Cube Provides

Component	Included	Description
`/mcp/*` endpoints	✅ Built-in	MCP endpoints for AI agents (all adapters)
`/cubejs-api/v1/explain`	✅ Built-in	Execution plan endpoint (all adapters)
Prompt templates	✅ Exported	`buildStep0Prompt`, `buildStep1Prompt`, etc.
Type definitions	✅ Exported	`Step0Result`, `Step1Result`, `AIExplainAnalysis`
`/api/ai/*` endpoints	❌ You build	Reference implementation provided

Reference Implementation

See our complete working implementation in the dev server:

ai-routes.ts - Complete AI endpoints (800+ lines)
app.ts - How to mount AI routes with middleware

Available AI Features

Natural Language Query Generation

Convert questions like “Show me sales by region this quarter” into properly structured semantic queries. The multi-stage generation pipeline fetches actual dimension values from your database, ensuring accurate filter values.

Learn more about Query Generation →

Query Execution Analysis

Analyze query execution plans with AI-powered recommendations. Get insights into:

Sequential scans that might benefit from indexes
Missing index opportunities
Query optimization suggestions
Performance assessments (good / warning / critical)

Learn more about Query Analysis →

Prerequisites

To use AI features, you need:

Google Gemini API Key - Get one free at Google AI Studio
Server-side AI Routes - AI endpoints must be implemented in your server (not included in adapters)

Architecture Overview

┌────────────────────────────────────────────────────────────────────┐
│                        Your Application                             │
├────────────────────────────────────────────────────────────────────┤
│                                                                    │
│  ┌─────────────────┐  ┌──────────────────┐  ┌───────────────────┐ │
│  │  Cube API       │  │  MCP Server      │  │  AI Routes        │ │
│  │  /cubejs-api/v1 │  │  /mcp (built-in) │  │  /api/ai (custom) │ │
│  │                 │  │                  │  │                   │ │
│  │  • /load        │  │  • meta          │  │  • /generate      │ │
│  │  • /meta        │  │  • discover      │  │  • /explain/...   │ │
│  │  • /explain     │  │  • validate      │  │  • /health        │ │
│  │  • /batch       │  │  • load          │  │                   │ │
│  └─────────────────┘  └──────────────────┘  └───────────────────┘ │
│           │                    │                     │             │
│           └────────────────────┴─────────────────────┘             │
│                                │                                   │
│                       ┌────────▼────────┐                          │
│                       │ Semantic Layer  │                          │
│                       │ + Security Ctx  │                          │
│                       └─────────────────┘                          │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

  ┌─────────────────┐        ┌──────────────────┐
  │   MCP Endpoints │        │  Custom AI       │
  │   (Built-in)    │        │  (You implement) │
  ├─────────────────┤        ├──────────────────┤
  │ • No LLM needed │        │ • LLM required   │
  │ • Zero config   │        │ • Customizable   │
  │ • Server-side   │        │ • Provider choice│
  │   NLU           │        │ • Full control   │
  └─────────────────┘        └──────────────────┘

Security Model

AI features respect your security context at every step:

Dimension value lookups use your security context - users only see values from their tenant’s data
Generated queries are executed with the same security context
Input validation (Step 0) rejects injection attempts, off-topic requests, and malicious prompts
Rate limiting prevents abuse of server API keys

Quick Start

Set up AI endpoints in your server
Configure the client to use AI features:

<CubeProvider
  apiOptions={{ apiUrl: '/cubejs-api/v1' }}
  features={{
    enableAI: true,
    aiEndpoint: '/api/ai/generate'
  }}
>
  <AnalysisBuilder />
</CubeProvider>