Skip to content

v1.0.x

Version 1.0.0: First stable release with full Reader, Splitter, Embedding and Vision model support.

v1.0.1

Add KeywordSplitter to split by regex patterns or specific keywords.

Features

Add new Splitter: KeywordSplitter. This Splitter allows to chunk based on regular expressions and patterns.

Documentation

Update documentation server to provide more examples

Developer features

Update pre-commit hooks to sync dependencies when executing the tests.

v1.0.0

Features

  • Consolidated all features introduced in v0.x series into a stable API.
  • Readers: VanillaReader, MarkItDownReader, DoclingReader with support for multiple formats (text, Office, JSON/YAML, images, HTML, etc.).
  • Splitters: character, word, sentence, paragraph, recursive, token, paged, row/column, JSON, semantic, HTML tag, header, and code splitting strategies.
  • Models: support for multimodal Vision-Language Models (OpenAI, Azure, Grok, HuggingFace, Gemini, Claude).
  • Embeddings: OpenAI, Azure, HuggingFace, Gemini, Claude (via Voyage) supported.

Developer features

  • Optional extras system: install lightweight core by default, extend with markitdown, docling, multimodal, azure, or all.
  • CI/CD pipeline, PyPI release, and pre-commit checks in place.

Documentation

Extensive documentation with API reference, examples, and architecture diagrams.

New improvements + Bug fixes

  • Fix: NLTK tokenizers in TokenSplitter are now correct base tokenizer when using nltk tokenizers.
  • Fix: RecursiveJSONSplitter could not produce outputs since it did not validate correct data type.
  • Now the examples are based on real Jupyter Notebooks executions to ensure that the behavior is the expected one.
  • Added the Notebooks which are used as examples in the notebooks section.
  • Update clean instruction with poe.
  • New constans have been defined.
  • Add new class to transform HTML to Markdown.

v0.6.x

[!IMPORTANT] Breaking Change! Version v0.6.0

Dependencies are now split into core (installed by default) and optional extras for heavy or specialized features. - Example: to use MarkItDown and Docling readers, install with: 

pip install "splitter-mr[markitdown,docling]"
- To install all optional features: 
pip install "splitter-mr[all]"

  • This change reduces install time and keeps core installs lightweight.

v0.6.5

Hotfix: dependency isolation was not guaranteed.

Features

  • Add a util class to convert HTML to Markdown content.
  • Improve Header Splitter to always return its content in markdown format.
  • Add the option to return text in markdown format for HTMLTagSplitter.
  • Add the option to batch content when using HTMLTagSplitter: If batch=True, it returns the chunks grouped by tags up to the numbers of characters specified by chunk_size. If False, it will return one register per tag.

Bug fixes

  • Dependency isolation was not guaranteed: implement safe lazy imports in all the __init__ methods.
  • Raise test coverage up to 90%.

0.6.4

Version 0.6.4:

SplitterMR now supports Anthropic Claude as a backend for both embedding (via Voyage AI) and vision models.

Features

  • Add new Vision Model: Claude Anthropic models.
  • Add new Embedding Model: Voyage Anthropic models.

Documentation

  • Change font type to Documentation server.
  • Update API reference guide with new links and resources.

v0.6.3

Version 0.6.3: SplitterMR now supports Gemini as a backend for both embedding and vision models.

To use HuggingFace, Gemini, Claude or Grok models, you must install SplitterMR with the multimodal extra:

pip install "splitter-mr[multimodal]"

Features

  • Add GeminiVisionModel class to Vision models.
  • Add GeminiEmbedding class to embedding models.
  • Apply lazy import strategy to classes which require extras to be installed (e.g., docling, markitdown, etc.).

Documentation

  • Update documentation.

v0.6.2

Version 0.6.2: SplitterMR now supports HuggingFace as a backend for both embedding and vision models:

  • HuggingFaceEmbedding: Use any Sentence Transformers model (local or from Hugging Face Hub) for fast, local, or cloud embeddings.
  • HuggingFaceVisionModel: Leverage Hugging Face’s vision-language models for image-to-text and image captioning.

To use HuggingFace, Gemini, Claude or Grok models, you must install SplitterMR with the multimodal extra:

pip install "splitter-mr[multimodal]"

Add HuggingFace Model and Embedding support.

Features

  • Add HuggingFaceVisionModel class. Note that the support is limited until now.
  • Add HuggingFaceEmbedding class.

Documentation

  • Add HuggingFaceVisionModel to documentation.
  • Update architecture diagram.
  • Update README.md.

Developer features

  • Add new dependencies to multimodal group.

v0.6.1

Add Grok Vision Model.

Version 0.6.1: SplitterMR now supports GrokVisionModel. See documentation here.

To use HuggingFace, Gemini, Claude or Grok models, you must install SplitterMR with the multimodal extra:

``bash pip install "splitter-mr[multimodal]"

Features

  • Add GrokVisionModel.
  • Redefine constants.
  • Add new tests.

Documentation

  • Add GrokVisionModel documentation.
  • Fix format bugs.
  • Add new documentation in Readers about how to install necessary dependencies.
  • Add plugin to read formulas appropiately.

v0.6.0

Divide library into sub packages.

Features

  • Divide the library into sub-modules.

Developer features

  • Add new steps to Dockerfile images.
  • Change requirements.txt to don't save editable builds as dependencies.
  • Change how the splitter_mr library is installed within Dockerfiles.
  • Lighten the weight of the library by making some dependencies optional.
  • Change how poe test is executed.

Documentation

  • Fix Embedding models not showing on Developer Guide overview page.

v0.5.x

[!IMPORTANT] New version v0.5.0

  • Add embedding models to encode the text into distributed vectorized representations. See documentation here.
  • Add support for chunking files based on semantic similarity between sentences. See documentation here.

v0.5.0

Add SemanticSplitter first implementation

Features

  • Add embedding module.
  • Add AzureOpenAI embeddings.
  • Add OpenAI embeddings.
  • Add BaseEmbeddings, to create your own class.
  • Add SemanticSplitter class.

Fixes

  • Fix SentenceSplitter class to be more robust and flexible (separators can be customized using regex pattern).

Developer features

  • Update tests.

Documentation

  • Update documentation with new embedding module.
  • Fix some format errors in Documentation server.
  • Add new example documentation page for SemanticSplitter.

v0.4.x

v0.4.0

PagedSplitter full implementation

[!IMPORTANT] New version v0.4.0

Add support for reading files and splitting them by pages using PageSplitter. Add support to read more files with VanillaReader.

➡️ See documentation.

Features

  • Add support to read a PDF by pages using MarkItDownReader without LLM.
  • Add method to read xlsx, pptx, docx files using VanillaReader.
  • Add method to read several image formats using VanillaReader.
  • Add support to read excel and parquet files using different engines in VanillaReader.
  • Add support to analyze content in several file types using AzureOpenAI and OpenAI models.

Documentation

  • Update documentation.
  • Fix some hyperlinks in README.

v0.3.x

v0.3.3

Features

  • Add a method to convert variables to a ReaderOutput object.
  • Add a page_placeholder attribute to the ReaderOutput object to distinguish when a file has been read by pages and which placeholder is.
  • Add an splitter method which split by pages for supported documents: PagedSplitter.

Developer features

  • Refactor the VanillaReader class to be more decoupled.

Documentation

  • Update examples in documentation server.

v0.3.2

Features

  • Add Pydantic models to validate inputs and outputs for BaseReader and BaseSplitter objects.
  • Refactor models to modularize into constants, pydantic models and prompts.

Documentation

  • Update README.md to handle notes and warnings.

v0.3.1

Features

  • Add support to read and scan PDF by pages for all the readers, using the parameter scan_pdf_images = True.
  • Add support to use different placeholders for images in Vanilla and Docling Readers.
  • Add support to split by pages for PDFs.
  • Add three different pipelines to DoclingReader to process the document as PageImages, using VLM to provide image captioning and regularly.
  • Add three different pipelines to VanillaReader to process the document as PageImages, using VLM to provide image captioning and regularly.

Bugs

  • Change how the arguments are passed to every Reader to enhance robutsness.
  • Add new test cases.

Documentation

  • Update examples.
  • Change MkDocs server to support both light and dark modes.

v0.3.0

[!IMPORTANT] Vision Language Model (VLM) support!

You can now use vision-capable models (OpenAI Vision, Azure OpenAI Vision) to extract image descriptions and OCR text during file reading. Pass a VLM model to any Reader class via the model parameter.

➡️ See documentation.

Features

  • Implement AzureOpenAI and OpenAI Vision Models to analyze graphical resources in PDF files.
  • Add support to read PDF files to VanillaReader using PDFPlumber.

Documentation

  • Update examples.
  • Add new examples to documentation.
  • Add Reading methods with PDF documentation.
  • Add information about implementing VLMs in your reading pipeline.
  • Change file names on data folder to be more descriptive.
  • Update README.md and CHANGELOG.

Fixes

  • Update tests.
  • Update docstrings.
  • Update TokenSplitter to raise Exceptions if no valid models are provided.
  • Update TokenSplitter to take as default argument a valid tiktoken model.
  • Change HTMLTagSplitter to take the headers if a table is provided.
  • Change HeaderSplitter to preserve headers in chunks.

v0.2.x

v0.2.2

Features

  • Implement TokenSplitter: split text into chunks by number of tokens, using selectable tokenizers (tiktoken, spacy, nltk).
  • MarkItDownReader now supports more file extensions: PDFs, audio, etc.

Fixes

  • HTMLTagSplitter does not correctly chunking the document as desired.
  • Change docstring documentation.

Documentation

  • Updated Splitter strategies documentation to include TokenSplitter.
  • Expanded example scripts and test scripts for end-to-end manual and automated verification of all Splitter strategies.
  • New examples in documentation server for HTMLTagSplitter.

Developer Features

  • Remove Pipfile and Pipfile.lock.
  • Update to poethepoet as task runner tool

Fixes

v0.2.1

Features

Implement CodeSplitter.

Fixes

  • Change docstring for BaseSplitter to update with current parameters.
  • Some minor warnings in documentations when deploying Mkdocs server.

v0.2.0

[!IMPORTANT] Breaking change!

  • All Readers now return ReaderOutput dataclass objects.
  • All Splitters now return SplitterOutput dataclass objects.

You must access fields using dot notation (e.g., result.text, result.chunks), not dictionary keys.

Features

New splitting strategy: RowColumnSplitter for flexible splitting of tabular data.

New reader_method attribute in output dataclasses.

Migration

Update all code/tests to use attribute access for results from Readers and Splitters.

Use .to_dict() on output if a dictionary is required.

Update any custom splitter/reader implementations to use the new output dataclasses.

v0.1.x

v0.1.3

Features

  • Add a new splitting strategy: RowColumnSplitter.

Fixes

  • Change Readers to properly handle JSON files.

Documentation

  • Update documentation.

v0.1.2

Features

  • Now VanillaReader can read from multiple sources: URL, text, file_path and dictionaries.

Fixes

  • By default, the document_id was None instead of and uuid in VanillaReader.
  • Some name changes for splitter_method attribute in SplitterOutput method.

Developer features

  • Extend CI/CD lifecycle. Now it uses Dockerfile to check tests and deploy docs.
  • Automate versioning for the Python project.
  • The project has been published to PyPI.org. New versions will be deployed using CI/CD script.
  • requirements.txt has been added in the root of the project.
  • A new stage in pre-commit has been introduced for generating the requirements.txt file.
  • Makefile extended: new commands to serve mkdocs. Now make clean remove more temporary files.

Documentation

  • Update documentation in examples for the Documentation server.
  • Documentation server now can be served using Make.

v0.1.1

Some bug fixes in HeaderSplitter and RecursiveCharacterSplitter, and documentation updates.

Bug fixes

  • chunk_overlap (between 0 and 1) was not working in the split method from RecursiveCharacterSplitter.
  • Some markdown code was not properly formatted in README.md.
  • Reformat examples from docstring documentation in every Reader and Splitter classes.
  • HeaderSplitter was not properly handling the headers in some markdown and HTML files.

Documentation

  • Some examples have been provided in the documentation (docs/, and in the documentation server).
  • New examples in docstrings.

v0.1.0

First version of the project

Functional features

  • Add first readers, VanillaReader: reader which reads the files and format them into a string.
  • DoclingReader: reader which uses the docling package to read the files.
  • MarkItDownReader: reader which uses the markitdown package to read the files.
  • Add first splitters: CharacterSplitter, RecursiveCharacterSplitter, WordSplitter, SentenceSplitter, ParagraphSplitter, HTMLTagSplitter, RecursiveJSONSplitter, HeaderSplitter:
  • The package can be installed using pip.
  • README.md has been updated.
  • Tests cases for main functionalities are available.
  • Some data has been added for testing purposes.
  • A documentation server is deployed with up-to-date information.

Developer features

  • Update pyproject.toml project information.
  • Add pre-commit configurations (flake8, check commit messages, run test coverage, and update documentation).
  • Add first Makefile commands (focused on developers):
  • make help: Provide a list with all the Make commands.
  • make clean: Clean temporal files and cache
  • make shell: Run a uv shell.
  • make install: Install uv CLI and pre-commit.
  • make precommit: Install pre-commit hooks.
  • make format: Run pyupgrade, isort, black, and flake8 for code style.