NTTP - Natural Text Transfer Protocol

Query databases with natural language using an LLM

npx nttp setup
npx nttp query "show me 5 users"

Quick Start

1. Install

npm install nttp

2. Setup

Interactive (recommended):

npx nttp setup

Non-interactive (for agents/CI):

npx nttp setup --non-interactive \
  --database-type=pg \
  --database-url=postgresql://user:pass@localhost:5432/db \
  --llm-provider=anthropic \
  --llm-api-key=sk-ant-...

3. Query

CLI:

npx nttp query "show active users"
npx nttp query "count pending orders"
npx nttp query "top 10 products by price"

Code:

import { NTTP } from 'nttp';

const nttp = await NTTP.fromEnv();
const result = await nttp.query("show active users");
console.log(result.data);
await nttp.close();

How It Works

3-layer caching system optimizes cost and performance:

L1: EXACT       Hash match         $0        <1ms (in-memory) or ~5ms (Redis)
L2: SEMANTIC    Embedding match    $0.0001   80ms
L3: LLM         Claude/GPT         $0.01     2-3s

Most queries hit L1 or L2. Only novel queries reach the LLM.

Features

• Natural Language Queries - "show active users" → SELECT * FROM users WHERE status = 'active'
• Auto Error Correction - LLM automatically fixes failed SQL (up to 3 attempts)
• 3-Layer Caching - Exact, semantic, and LLM-generated query caching
• Multi-LLM Support - Claude, GPT-4, Cohere, Mistral, Gemini
• Multi-Database - PostgreSQL, MySQL, SQLite, SQL Server
• Redis Persistence - Optional cache persistence across restarts
• Type-Safe - Full TypeScript support
• CLI + SDK - Use from command line or code

Documentation

• Quick Reference: npx nttp docs [topic]
• Full Guides: /docs

Guides

• API Reference - Complete API documentation
• Caching System - 3-layer cache deep dive
• Configuration - All config options
• LLM Models - Model selection guide
• Examples - Usage examples
• Production - Deployment best practices
• Troubleshooting - Common issues

Example Queries

// Simple
await nttp.query("show users");
await nttp.query("list products");

// Filtered
await nttp.query("active users from California");
await nttp.query("products under $50");

// Aggregations
await nttp.query("count pending orders");
await nttp.query("total revenue by category");

// Complex
await nttp.query("top 10 products by revenue");
await nttp.query("users who joined this year");

Configuration

Environment Variables (.env)

# Database
DATABASE_TYPE=pg
DATABASE_URL=postgresql://user:pass@localhost:5432/mydb

# LLM
LLM_PROVIDER=anthropic
LLM_MODEL=claude-sonnet-4-5-20250929
ANTHROPIC_API_KEY=sk-ant-...

# Cache (optional but recommended)
REDIS_URL=redis://localhost:6379
OPENAI_API_KEY=sk-...  # For L2 semantic cache

Programmatic

const nttp = new NTTP({
  database: {
    client: 'pg',
    connection: 'postgresql://user:pass@localhost:5432/mydb'
  },
  llm: {
    provider: 'anthropic',
    model: 'claude-sonnet-4-5-20250929',
    apiKey: process.env.ANTHROPIC_API_KEY
  },
  cache: {
    redis: { url: 'redis://localhost:6379' },
    l2: { enabled: true, provider: 'openai', apiKey: process.env.OPENAI_API_KEY }
  }
});

await nttp.init();

See Configuration Guide for all options.

Performance

• L1 Cache (Exact): <1ms (in-memory) or ~5ms (Redis) - $0
• L2 Cache (Semantic): ~75ms - ~$0.0001
• L3 (LLM): 2-3s - ~$0.01
• Throughput: >10,000 req/s (cached)

Cost Savings: With caching, 90%+ cost reduction after warm-up.

Supported Databases

• PostgreSQL - Recommended for production
• MySQL - Widely supported
• SQLite - Perfect for development
• SQL Server - Enterprise-ready

Supported LLMs

• Anthropic Claude (recommended) - Best SQL generation
• OpenAI GPT - Fast and reliable
• Cohere - Enterprise support
• Mistral - Open-source preference
• Google Gemini - Multimodal capabilities

See Model Selection Guide for detailed comparison.

CLI Commands

# Setup wizard
npx nttp setup

# Query database
npx nttp query "your question"
npx nttp query "show users" --format json

# Documentation
npx nttp docs                    # Show all docs
npx nttp docs redis              # Search for "redis"
npx nttp docs "semantic cache"   # Multi-word search

API Overview

import { NTTP } from 'nttp';

// Initialize from environment variables
const nttp = await NTTP.fromEnv();

// Execute query
const result = await nttp.query("show active users");
console.log(result.data);        // Query results
console.log(result.cacheHit);    // true/false
console.log(result.meta);        // Cache metadata

// Explain query (without executing)
const explanation = await nttp.explain("show users");
console.log(explanation.sql);    // Generated SQL

// Database inspection
const tables = await nttp.getTables();
const schema = await nttp.getTableSchema('users');

// Cache management
const stats = await nttp.getCacheStats();
await nttp.pinSchema(schemaId);

// Clean up
await nttp.close();

See API Reference for complete documentation.

Error Handling

import { IntentParseError, SQLGenerationError, SQLExecutionError } from 'nttp';

try {
  const result = await nttp.query("your query");
} catch (error) {
  if (error instanceof IntentParseError) {
    console.error('Could not understand query');
    console.log('Suggestions:', error.suggestions);
  } else if (error instanceof SQLGenerationError) {
    console.error('Could not generate SQL');
  } else if (error instanceof SQLExecutionError) {
    console.error('Query execution failed');
  }
}

All errors include helpful suggestions. See Troubleshooting Guide.

TypeScript

Full TypeScript support with exported types:

import type {
  NTTPConfig,
  QueryResult,
  Intent,
  SchemaDefinition,
  CacheStats
} from 'nttp';

Examples

• Basic Usage
• Express Integration
• Next.js Integration
• CLI Tools

Links

• GitHub
• npm
• Issues
• Documentation

License

MIT