LLM-OCR

A simple OCR SDK that uses AI models to extract text from images and return formatted markdown.

Features

• [x] Support for multiple AI providers (Google Gemini, Mistral)
• [x] Local and remote image processing
• [x] Streaming and non-streaming responses
• [x] Base64 image encoding
• [x] Markdown formatted output
• [ ] additional provider support with models
• [ ] additional output formats (JSON)
• [ ] support for pdf files
• [ ] support for Multi-page PDF files

Installation

npm install llm-ocr
# or
yarn add llm-ocr
# or
pnpm add llm-ocr
# or
bun add llm-ocr

Environment Variables

Create a .env file and add your API keys:

GOOGLE_API_KEY=your_google_api_key
MISTRAL_API_KEY=your_mistral_api_key

Usage

Basic Example

import { ocr } from 'llm-ocr';

// For local image
const result = await ocr({
  filePath: './path/to/image.jpg',
  modelID: 'gemini-1.5-flash',
  provider: 'google',
  stream: false,
  // systemPrompt: 'What is the text in the image?', // Optional
});

// For remote image
const result = await ocr({
  filePath: 'https://example.com/image.jpg',
  modelID: 'pixtral-large-latest',
  provider: 'mistral',
  stream: false,
});

Available Models

Google Models:

• gemini-1.5-flash fast but less accurate
• gemini-1.5-flash-8b fast but less accurate
• gemini-1.5-pro accurate but slow

Mistral Models:

• pixtral-12b-2409 fast but less accurate
• pixtral-large-latest accurate but slow

Utility Functions

import { encodeImage, isRemoteFile, downloadImageAndEncode } from 'llm-ocr';

// Encode local image to base64
const base64Image = encodeImage('./path/to/image.jpg');

// Check if file is remote
const isRemote = isRemoteFile('https://example.com/image.jpg');

// Download and encode remote image
const encodedRemoteImage = await downloadImageAndEncode(
  'https://example.com/image.jpg',
);

License

MIT © Vivek Patel