Drizzle LLM

Build-time SQL generation using LLM with natural language queries.

Installation

npm install drizzle-llm

Quick Start

1. Configure

// drizzle-llm.config.ts
export default {
  provider: {
    type: 'openai',
    apiKey: process.env.OPENAI_API_KEY,
  },
  paths: {
    schema: './src/schema',
    queries: './src/**/*.ts',
  }
};

2. Write Queries

// src/user.ts
import { llm } from "drizzle-llm";

export async function getUser(db: any, id: string): Promise<User[]> {
  return db.all<User>(llm`ユーザーを${id}で取得する`);
}

export async function findUserName(db: any, id: string): Promise<string | null> {
  return db.get<string>(llm`Find user name by id ${id}`);
}

💡 Tip: Be explicit with types and methods for better SQL generation:

• Use db.get() for single results, db.all() for multiple
• Add TypeScript return types: Promise<User[]>, Promise<string | null>
• Use generic types: db.get<string>(), db.all<User>()

3. Build

npx drizzle-llm build

This generates user.sql:

-- 4a5f9c682c34d6ec808677691ea62d47
-- ユーザーを${0}で取得する
SELECT * FROM user WHERE id = $1

How It Works

1. Scan: Finds llm\...`` calls in your code
2. Generate: Uses LLM to create SQL from natural language
3. Cache: Stores results to avoid redundant API calls
4. Runtime: Loads pre-generated SQL at runtime

Environment Variables

OPENAI_API_KEY=sk-...
DRIZZLE_LLM_AUTO_APPROVE=true  # Skip confirmation prompts

CLI Commands

$ npx drizzle-llm build    # Generate SQL queries

Configuration

interface DrizzleLLMConfig {
  provider: {
    type: 'openai' | 'anthropic';
    apiKey: string;
    model?: string;
  };
  paths: {
    schema: string;
    queries: string | string[];
  };
  cache?: {
    enabled: boolean;
    directory: string;
  };
}

Author

• github/iwamatsu0430

License

MIT © iwamatsu0430