Getting started

This guide will help you get the most out of the Mindee Ruby client library to easily extract data from your documents.



The following Ruby versions are tested and supported: 2.6, 2.7, 3.0, 3.1, 3.2

Standard Installation

To quickly get started with the Ruby OCR SDK, Install by adding this line to your application's Gemfile:

gem 'mindee'

And then execute:

bundle install

Or you can install it like this:

gem install mindee

Finally, Ruby away!

Development Installation

If you'll be modifying the source code, you'll need to install the required libraries to get started.

We recommend using Bundler.

  1. First clone the repo.
git clone [email protected]:mindee/mindee-api-ruby.git
  1. Navigate to the cloned directory and install all required libraries.
cd mindee-api-ruby
bundle install

Updating the Library

It is important to always check the version of the Mindee OCR SDK you are using, as new and updated
features won’t work on older versions.

To get the latest version of your OCR SDK:

gem install mindee

To install a specific version of Mindee:

gem install mindee@<version>


Using Mindee's APIs can be broken down into the following steps:

  1. Initialize a Client
  2. Load a File
  3. Send the File to Mindee's API
  4. Process the Result in some way

Let's take a deep dive into how this works.

Initializing the Client

The Client automatically connects to the default endpoints for each product (or creates one with given parameters for Custom APIs).

The Client requires your API key.

You can either pass these directly to the constructor or through environment variables.

Pass the API key directly

# Init a new client and passing the key directly
mindee_client = 'my-api-key')

Set the API key in the environment

API keys should be set as environment variables, especially for any production deployment.

The following environment variable will set the global API key:


Then in your code:

# Init a new client without an API key
mindee_client =

Setting the Request Timeout

The request timeout can be set using an environment variable:


Loading a Document File

Before being able to send a document to the API, it must first be loaded.

You don't need to worry about different MIME types, the library will take care of handling
all supported types automatically.

Once a document is loaded, interacting with it is done in exactly the same way, regardless
of how it was loaded.

There are a few different ways of loading a document file, depending on your use case:


Load from a file directly from disk. Requires an absolute path, as a string.

# Init a new client
mindee_client = 'my-api-key')

# Load a file from disk
input_source = mindee_client.source_from_path('/path/to/the/file.ext')

# Parse the file
result = mindee_client.parse(
# Print a full summary of the parsed data in RST format
puts result.document

File Object

A normal Ruby file object with a path. Must be in binary mode.

Note: The original filename is required when calling the method., 'rb') do |fo|
  input_source = mindee_client.source_from_file(fo, "invoice.jpg")

result = mindee_client.parse(


Load file contents from a base64-encoded string.

Note: The original filename is required when calling the method.


input_source = mindee_client.source_from_b64string(b64_string, "receipt.jpg")

result = mindee_client.parse(


Requires raw bytes.

Note: The original filename is required when calling the method.

raw_bytes = b"%PDF-1.3\n%\xbf\xf7\xa2\xfe\n1 0 ob..."
input_source = mindee_client.source_from_bytes(raw_bytes, "invoice.pdf")

result = mindee_client.parse(


Requires an url as a String.

Note: the url must start with https://.

input_source = mindee_client.source_from_url("")

result = mindee_client.parse(

Sending a File

To send a file to the API, we need to specify how to process the document.
This will determine which API endpoint is used and how the API return will be handled internally by the library.

More specifically, we need to set a Mindee::Product class as the first parameter of the create_endpoint method.

This is because the Endpoint's urls will be set according to it

Each document type available in the library has its corresponding class, which inherit from the base Mindee::Parsing::Common::Predict class.
This is detailed in each document-specific guide.

Off-the-Shelf Documents

Simply setting the correct class is enough:

result = mindee_client.parse(

Custom Documents

For custom documents, the endpoint to use must also be set, and it must take in an endpoint_name:

endpoint = mindee_client.create_endpoint(endpoint_name: 'wnine')

result = mindee_client.parse(
  endpoint: endpoint

This is because the CustomV1 class is enough to handle the return processing, but the actual endpoint needs to be specified.

Process the Result

The response object is common to all documents, including custom documents. The main properties are:

  • id — Mindee ID of the document
  • name — Filename sent to the API
  • inferenceInference


Regroups the predictions at the page level, as well as predictions for the entire document.

Document level prediction

The prediction attribute is a Prediction object specific to the type of document being processed.
It contains the data extracted from the entire document, all pages combined.

It's possible to have the same field in various pages, but at the document level,
only the highest confidence field data will be shown (this is all done automatically at the API level).

# as an object, complete
pp result.document.inference.prediction

# as a string, summary in RST format
puts result.document.inference.prediction

Page level prediction

The pages attribute is a list of Prediction objects.

Each page element contains the data extracted for a particular page of the document.
The order of the elements in the array matches the order of the pages in the document.

All response objects have this property, regardless of the number of pages.
Single page documents will have a single entry.

Iteration is done like any Ruby array:

response.document.inference.pages.each do |page|
    # as an object, complete
    pp page.prediction

    # as a string, summary in RST format
    puts page.prediction

Page Orientation

The orientation field is only available at the page level as it describes whether the page image should be rotated to be upright.

If the page requires rotation for correct display, the orientation field gives a prediction among these 3 possible outputs:

  • 0 degrees: the page is already upright
  • 90 degrees: the page must be rotated clockwise to be upright
  • 270 degrees: the page must be rotated counterclockwise to be upright
response.document.inference.pages.each do |page|
  puts page.orientation.value

🧪 Experimental Features

PDF repair

Some PDF files might appear fine on your computer, but can be rejected by the server.
This experimental feature attempts to fix the file's header information before sending it to the server.


Warning: This feature copies your file and then alters it. The original file will be left alone, but the copy might get partially corrupted, and improperly parsed as a result. Use at your own discretion.

To enable it, simply set the fix_pdf flag to true during source creation:

input_source = mindee_client.source_from_file(input_file, "name-of-my-file.ext", fix_pdf: true)

Note: This only works for local files, files sent by URL will not be processed.


Join our Slack