pageslice 📄⚡

Zero-glue document parser & AI extraction toolkit for Node.js & TypeScript.
Slice any PDF, DOCX, XLSX, TXT, or Markdown into raw text or guaranteed typed JSON using Zod schemas.

Why pageslice?

Every developer building document AI in JavaScript/TypeScript knows the pain:

• Installing pdf-parse, mammoth, and fighting node-gyp / C++ native build errors in Next.js, Vercel, and Docker.
• Writing 80+ lines of glue code to handle file formats, buffers, and clean whitespace.
• Dealing with LLMs returning broken JSON or missing schema fields.

pageslice eliminates all of that. One single package. Zero native C++ dependencies. Pure JavaScript. Works offline for text parsing, and supports OpenAI, Google Gemini, OpenRouter, or local Ollama for AI.

Key Features

• 📦 All-in-One: PDF, DOCX, XLSX/XLS (Excel), TXT, MD, CSV, JSON support out of the box.
• 🔓 0 API Keys Needed for Text: Extract text from PDF, Word, or Excel docs 100% offline with zero setup.
• ⚡ Zero Native C++ Dependencies: No node-gyp, canvas, or poppler. Runs cleanly on Next.js, Vercel Serverless, AWS Lambda, and Bun.
• 📊 Smart Excel → Markdown Tables: Converts spreadsheet sheets into clean markdown tables for optimal LLM comprehension and structured extraction.
• 🎯 Guaranteed Type-Safe JSON: Built-in Zod schema validation.
• 🔄 Self-Healing Retries: If an LLM returns invalid JSON or misses fields, pageslice automatically reprompts with the validation errors and fixes it.
• 🔑 Bring Your Own Key (BYOK): Supports OpenAI, OpenRouter (100+ models), Google Gemini, and any custom OpenAI-compatible endpoint.
• 🌐 Universal Inputs: Accepts file paths, Node Buffer, Uint8Array, web URL, and browser Blob/File.
• 🧩 Zod Re-exported: Use import { z } from 'zod' or import { z } from 'pageslice' — both work seamlessly.

Installation

npm install pageslice
# or
pnpm add pageslice
# or
bun add pageslice

⚡ Instant CLI Commands (Zero Code Required)

You don't even need to write a line of code to test or parse documents. You can run pageslice directly from your terminal using npx:

1. Run Live Diagnostic Test (See Real Files Parsed Live)

Runs built-in parsers on real PDF, DOCX, and XLSX files and prints the entire extracted text and markdown tables right on your screen:

npx pageslice test

2. Parse Any File Directly from Your Terminal

Extract and inspect text or tables from any local document in 50ms:

# Parse a PDF file
npx pageslice ./resume.pdf

# Parse an Excel spreadsheet into Markdown tables
npx pageslice ./sales.xlsx

# Parse a Word document
npx pageslice ./contract.docx

Code Quickstart

1. Plain Text Extraction (NO API Keys Required, 100% Offline)

If you just want the raw text from any PDF, Word doc, Excel spreadsheet, or text file:

import { parse } from 'pageslice';

// Works instantly, offline, zero API keys required!
const document = await parse('./contract.pdf'); // or .docx, .xlsx, .txt, .json

console.log(document.text);       // Full extracted text
console.log(document.pages);      // Array of text per page (or per sheet for Excel)
console.log(document.totalPages); // e.g. 5

Excel Parsing (Automatic Markdown Table Conversion)

Excel files are automatically converted into clean markdown tables — perfect for LLM comprehension:

import { parse } from 'pageslice';

const spreadsheet = await parse('./sales-data.xlsx');

console.log(spreadsheet.text);
// Output:
// ## Sheet: Q1 Sales
//
// | Product | Units | Revenue |
// | --- | --- | --- |
// | Widget A | 1200 | 45000 |
// | Widget B | 800 | 32000 |
//
// ## Sheet: Q2 Sales
// ...

console.log(spreadsheet.totalPages);  // Number of sheets
console.log(spreadsheet.metadata);    // { sheetNames: ['Q1 Sales', 'Q2 Sales'], totalSheets: 2 }

2. AI Q&A & Guaranteed Typed JSON (Zod)

When you want AI Q&A or structured JSON extraction:

import { doc, z } from 'pageslice'; // or import { z } from 'zod'

// 1. Load document (auto-detects format)
const document = await doc('./invoice.pdf', {
  provider: 'openrouter', // or 'openai' | 'gemini' | 'custom'
  apiKey: process.env.OPENROUTER_API_KEY,
});

// 2. Ask any freeform question
const answer = await document.ask('What are the payment terms?');
console.log(answer);

// 3. Extract guaranteed typed data using Zod
const { data } = await document.extract({
  prompt: 'Extract vendor details and totals',
  schema: z.object({
    vendorName: z.string(),
    invoiceNumber: z.string(),
    totalAmount: z.number(),
    lineItems: z.array(
      z.object({
        description: z.string(),
        price: z.number(),
      })
    ),
  }),
});

// 100% typed & autocompleted in TypeScript!
console.log(data.vendorName, data.totalAmount);

All AI Provider Setup Examples

1. OpenAI Setup

import { doc } from 'pageslice';

const document = await doc('./invoice.pdf', {
  provider: 'openai',
  apiKey: process.env.OPENAI_API_KEY,
  model: 'gpt-4o-mini', // default is 'gpt-4o-mini', can also use 'gpt-4o'
});

2. Google Gemini Setup

import { doc } from 'pageslice';

const document = await doc('./invoice.pdf', {
  provider: 'gemini',
  apiKey: process.env.GEMINI_API_KEY,
  model: 'gemini-1.5-flash', // or 'gemini-2.5-flash'
});

3. OpenRouter Setup (100+ Models: Claude 3.5 Sonnet, Llama 3, DeepSeek)

import { doc } from 'pageslice';

const document = await doc('./invoice.pdf', {
  provider: 'openrouter',
  apiKey: process.env.OPENROUTER_API_KEY,
  model: 'anthropic/claude-3.5-sonnet', // or 'google/gemini-2.5-flash'
});

4. Local Ollama / vLLM / Custom Setup (Free & Offline AI)

import { doc } from 'pageslice';

const document = await doc('./file.pdf', {
  provider: 'custom',
  baseURL: 'http://localhost:11434/v1', // local Ollama server
  model: 'llama3',
});

5. Environment Variable Auto-Detection (Zero Config)

If you put OPENROUTER_API_KEY, OPENAI_API_KEY, or GEMINI_API_KEY in your .env file, you don't even need to pass options:

import { doc } from 'pageslice';

// Automatically detects process.env.OPENROUTER_API_KEY or OPENAI_API_KEY!
const document = await doc('./file.pdf');
const answer = await document.ask('What is this file about?');

Universal Input Support

pageslice handles any input type without requiring manual conversions:

// Local file path
await doc('./contract.docx');

// Multer / Express / Next.js file upload Buffer
await doc(req.file.buffer, { format: 'pdf' });

// Web URL (automatically fetches and parses)
await doc('https://example.com/invoice.pdf');

// Browser / FormData Blob / File
await doc(fileBlob);

// Raw text string
await doc('Company policy notes...');

Backend Integration Examples

Next.js (App Router Route Handler)

// app/api/parse-doc/route.ts
import { NextResponse } from 'next/server';
import { doc, z } from 'pageslice';

export async function POST(req: Request) {
  const formData = await req.formData();
  const file = formData.get('document') as File;

  const buffer = Buffer.from(await file.arrayBuffer());

  const document = await doc(buffer, {
    provider: 'openrouter',
    apiKey: process.env.OPENROUTER_API_KEY,
  });

  const { data } = await document.extract({
    prompt: 'Extract document summary and metrics',
    schema: z.object({
      title: z.string(),
      summary: z.string(),
    }),
  });

  return NextResponse.json({ data });
}

Express.js (Node.js API with Multer)

import express from 'express';
import multer from 'multer';
import { doc, z } from 'pageslice';

const app = express();
const upload = multer({ storage: multer.memoryStorage() });

app.post('/api/parse', upload.single('document'), async (req, res) => {
  const document = await doc(req.file.buffer);

  const { data } = await document.extract({
    schema: z.object({ vendor: z.string(), amount: z.number() }),
  });

  res.json({ data });
});

Reusable Client (createDocClient)

Configure your provider once at the top of your app and use it across multiple API routes:

// lib/docClient.ts
import { createDocClient } from 'pageslice';

export const client = createDocClient({
  provider: 'openrouter',
  apiKey: process.env.OPENROUTER_API_KEY,
  model: 'anthropic/claude-3.5-sonnet',
});

// In any server action or route:
import { client } from '@/lib/docClient';
import { z } from 'zod';

export async function processUpload(fileBuffer: Buffer) {
  const document = await client.load(fileBuffer);
  
  return await document.extractData({
    schema: z.object({ title: z.string(), total: z.number() }),
  });
}

Comparison

| Feature | LangChain JS | Raw pdf-parse / mammoth | pageslice | | :--- | :--- | :--- | :--- | | npm packages required | 6 to 8 | 3 to 5 | Just 1 (pageslice) | | Native C++ build issues? | Frequently | Frequently | Zero (pure JS) | | Instant CLI testing | ❌ No | ❌ No | ✅ Yes (npx pageslice test) | | Direct terminal file parser | ❌ No | ❌ No | ✅ Yes (npx pageslice <file>) | | Text parsing offline (No API key) | ❌ No | ✅ Yes | ✅ Yes (parse()) | | Supported formats | Separate loaders per format | PDF or DOCX only | PDF, DOCX, XLSX, TXT, MD, CSV, JSON | | Excel → Markdown tables | ❌ No | ❌ No | ✅ Auto-converts sheets to tables | | Guaranteed Zod JSON | Verbose LCEL Chains | Manual | Built-in with Self-Healing | | Next.js & Vercel Ready | ⚠️ Cold starts | ⚠️ Node-gyp errors | ✅ Edge & Serverless Ready |

🧪 Run Test Suite

You can clone the repository and run the test suite:

npm test
# or
npm run pageslice:test

All tests run 100% offline in pure JavaScript without requiring any API keys or internet connection.

License

MIT © pageslice Contributors