Preprocess is an API service that splits various types of documents into optimal chunks of text for use in language model tasks. It divides documents into chunks that respect the layout and semantics of the original content, accounting for sections, paragraphs, lists, images, data tables, text tables, and slides.

We support the following formats:

• PDFs
• Microsoft Office documents (Word, PowerPoint, Excel)
• OpenOffice documents (ODS, ODT, ODP)
• HTML content (web pages, articles, emails)
• Plain text

To install the Python Preprocess library, use:

pip install pypreprocess

Alternatively, to add it as a dependency with Poetry:

poetry add pypreprocess
poetry install

Note: You need a Preprocess API Key to use the SDK. To obtain one, please contact support@preprocess.co.

Retrieve chunks from a file for use in your language model tasks:

from pypreprocess import Preprocess

# Initialize the SDK with a file
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")

# Chunk the file
preprocess.chunk()
preprocess.wait()

# Get the result
result = preprocess.result()
for chunk in result.data['chunks']:
    # Use the chunks

You can initialize the SDK in three different ways:

1- Passing a local filepath:

Use this when you want to chunk a local file:

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")

2- Passing a process_id:

When the chunking process starts, Preprocess generates a process_id that can be used to initialize the SDK later:

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, process_id="id_of_the_process")

3- Passing a PreprocessResponse Object:

When you need to store and reload the result of a chunking process later, you can use the PreprocessResponse object:

import json
from pypreprocess import Preprocess, PreprocessResponse
response = PreprocessResponse(**json.loads("The JSON result from a previous chunking process."))
preprocess = Preprocess(api_key=YOUR_API_KEY, process=response)

Preprocess offers several configuration options to tailor the chunking process to your needs.

Note: Preprocess attempts to output chunks with less than 512 tokens. Longer chunks may sometimes be produced to preserve content integrity. We are currently working to allow user-defined chunk lengths.

You can pass these parameters during SDK initialization:

preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file", merge=True, repeat_title=True, ...)
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file", options={"merge": True, "repeat_title": True, ...})

Or, set them later using the set_options method with a dict:

preprocess.set_options({"merge": True, "repeat_title": True, ...})
preprocess.set_options(merge=True, repeat_title=True, ...)

Note: if the parameter is present inside options dictionary, it will override the parameter passed in the function.

After initializing the SDK with a filepath, use the chunk() method to start chunking the file:

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")
response = preprocess.chunk()

The response contains the process_id and details about the API call's success.

The chunking process may take some time. You can wait for completion using the wait() method:

result = preprocess.wait()
print(result.data['chunks'])

In more complex workflows, store the process_id and retrieve the result later:

# Start chunking process
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")
preprocess.chunk()
process_id = preprocess.get_process_id()

# In a different flow
preprocess = Preprocess(api_key=YOUR_API_KEY, process_id=process_id)
result = preprocess.wait()
print(result.data['chunks'])

Alternatively, use the result() method to check if the process is complete:

result = preprocess.result()
if result.data['process']['status'] == "FINISHED": 
    print(result.data['chunks'])

Here are additional methods available in the SDK:

• set_filepath(path): Set the file path after initialization.
• set_process_id(id): Set the process_id parameter by ID.
• set_process(PreprocessResponse): Set the process_id using a PreprocessResponse object.
• set_options(dict): Set chunking options using a dictionary.
• to_json(): Return a JSON string representing the current object.
• get_process_id(): Retrieve the current process_id.
• get_filepath(): Retrieve the file path.
• get_options(): Retrieve the current chunking options.