nextstring

nextstring is a powerful TypeScript library that extends the native String prototype with AI-driven capabilities. Perform advanced operations on strings — extract data, answer questions, summarize, translate, classify, score, mask sensitive data, and more — all powered by a configurable AI provider.

Features

Method	Description
extractData(items)	Extract structured data from text with optional type parsing
question(q)	Answer questions using the string as context
summarise(words)	Summarize text to a target word count
checkIf(condition)	Evaluate a condition against the string's content
translate(language)	Translate text to a target language
rewrite(instructions)	Rewrite text following specific instructions
classifyText(categories)	Classify text into predefined categories
fill(context)	Fill {{placeholder}} templates using AI-inferred values
generate(instructions?)	Use the string as a prompt to generate content
score(rubrics)	Score text against rubrics, returns numeric scores
mask(rules)	Partially mask sensitive data (emails, phones, SSNs, etc.)
detectLanguage()	Detect the language of the given text

All methods accept an optional { model: string } options object as the last parameter to override the default model per call.

Installation

npm install nextstring

Quick Start

1. Initialize with a provider

import { initialise, OpenaiProvider } from "nextstring";

// OpenAI
const provider = new OpenaiProvider({ apiKey: "sk-..." });
initialise(provider);

// or Gemini
import { GeminiProvider } from "nextstring";
const provider = new GeminiProvider({ apiKey: "AI..." });
initialise(provider);

// or Claude
import { ClaudeProvider } from "nextstring";
const provider = new ClaudeProvider({ apiKey: "sk-ant-..." });
initialise(provider);

2. Use AI methods on any string

const answer = await "The Eiffel Tower is in Paris.".question("Where is the Eiffel Tower?");
// "Paris"

Providers

Provider	Default Model	Import
OpenAI	gpt-4o-mini	OpenaiProvider
Google Gemini	gemini-2.5-flash	GeminiProvider
Anthropic Claude	claude-sonnet-4-20250514	ClaudeProvider

Each provider accepts a globalPrompt parameter to prepend instructions to every request.

const provider = new OpenaiProvider(
  { apiKey: "sk-..." },
  "gpt-4o",           // model
  "Always be concise"  // globalPrompt
);

Examples

Extract Data

Extract structured fields with optional type parsing via parse:

const text = "John is 30, email: john@example.com";
const result = await text.extractData([
  { name: "email", description: "Email address" },
  { name: "age", description: "Age", parse: Number },
]);
// { email: "john@example.com", age: 30 }
// result.email → string, result.age → number (inferred from parse)

Summarize

const summary = await "A very long article about climate change...".summarise(20);

Check Conditions

const isSunny = await "The weather today is sunny and warm.".checkIf("Is it sunny?");
// true

Translate

const spanish = await "Hello, how are you?".translate("spanish");
// "Hola, ¿cómo estás?"

Rewrite

const formal = await "hey whats up".rewrite("Make it formal");
// "Hello, how are you doing?"

Classify Text

const category = await "My app keeps crashing".classifyText([
  { name: "support", description: "Technical support issues" },
  { name: "sales", description: "Sales inquiries" },
]);
// "support"

Fill Templates

const filled = await "Dear {{name}}, your order {{orderId}} is {{status}}.".fill({
  name: "Alice",
  order: { id: "ORD-123", shipped: true },
});
// "Dear Alice, your order ORD-123 is shipped."

Generate Content

const content = await "a product description for running shoes".generate();

const email = await "a professional email declining a meeting".generate(
  "Keep it under 50 words"
);

Score Text

Score text against custom rubrics (returns 0-10 for each):

const scores = await "The quick brown fox jumps over the lazy dog.".score([
  { id: "grammar", name: "Grammar", description: "Correct grammar and punctuation" },
  { id: "clarity", name: "Clarity", description: "Clear and easy to understand" },
]);
// { grammar: 9, clarity: 8 }

Mask Sensitive Data

Partially redact sensitive information while preserving format:

const masked = await "Contact john@example.com or call 555-123-4567".mask(["email", "phone"]);
// "Contact j***@e******.com or call ***-***-4567"

const masked2 = await "SSN: 123-45-6789".mask(["ssn"]);
// "SSN: ***-**-6789"

Detect Language

const lang = await "Bonjour le monde".detectLanguage();
// "French"

const lang2 = await "こんにちは世界".detectLanguage();
// "Japanese"

Per-Call Model Override

Any method accepts an optional model override as the last argument:

const answer = await "some text".question("What is this?", { model: "gpt-4o" });
const summary = await "long text".summarise(10, { model: "claude-sonnet-4-20250514" });

Full Scenario: AI-Powered Customer Support

const message = "Bonjour, je suis Alice. La commande #12345 doit être remboursée. Mon email est alice@test.com";

// Detect language for routing
const language = await message.detectLanguage();
// "French"

// Extract structured data
const data = await message.extractData([
  { name: "name", description: "Customer name" },
  { name: "orderID", description: "Order ID" },
  { name: "email", description: "Email address" },
]);
// { name: "Alice", orderID: "12345", email: "alice@test.com" }

// Translate to English for internal processing
const english = await message.translate("english");
// "Hello, I'm Alice. Order #12345 needs a refund. My email is alice@test.com"

// Understand intent
const intent = await english.classifyText([
  { name: "refund", description: "Refund requests" },
  { name: "tracking", description: "Order tracking" },
  { name: "general", description: "General inquiries" },
]);
// "refund"

// Mask PII for logging
const safe = await english.mask(["email", "name"]);
// "Hello, I'm [NAME]. Order #12345 needs a refund. My email is a****@t***.com"

// Generate a reply and translate back to the customer's language
const reply = await "a polite refund confirmation for order #12345".generate();
const localizedReply = await reply.translate(language);

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

License

This project is licensed under the MIT License.