feat: implement chonkie integration with four chunkers

[yugborana]

Related Issues

• fixes Add new Chonkie integration with RecursiveChunker, SemanticChunker and others #2237

Proposed Changes:

Implemented the Chonkie integration for Haystack by building four modular preprocessing components: ChonkieSemanticChunker, ChonkieRecursiveChunker, ChonkieTokenChunker, and ChonkieSentenceChunker.

Each component initializes its respective Chonkie engine with parameters (like thresholds, chunk_overlap, and custom delimiter rules). When executed in a pipeline (run()), the components delegate the input text to the underlying Chonkie algorithm. Finally, they package the resulting chunks back into new Haystack Document objects—preserving the original metadata while injecting new, precise annotations like start_index, end_index, and token_count.

How did you test it?

Unit Tests: Built 16 tests covering initialization, to_dict/from_dict serialization, and run() logic for all 4 chunkers. Ran hatch run test:unit and achieved a 100% pass rate.
Integration Tests: N/A (Chonkie runs locally and doesn't rely on external databases or cloud APIs).
Manual Verification: Successfully executed live_test.py on real models (minishlab/potion-base-32M) to confirm dynamic semantic splits, token overlaps, and custom delimiter extraction without using mocks.
Instructions for Manual Tests:
cd integrations/chonkie
Run unit tests: hatch run test:unit
Run the live demonstration: hatch run python live_test.py

Notes for the reviewer

Checklist

yes - I have read the contributors guidelines and the code of conduct

• I have updated the related issue with new insights and changes
  yes - I added unit tests and updated the docstrings
  yes - I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	chonkie@​1.6.4

View full report

[sjrl]

Hey @yugborana thanks for opening the contribution! I left an in depth review for one of the chunkers if you could also apply the same comments to other chunkers where appropriate that would be very appreciated!

[yugborana]

@sjrl thanks for the review. It was pretty fast. I will take forward from here...

[sjrl]

Thanks @yugborana!

Also I discussed with the team and we think to be more consistent with our existing component name we should rename the Chonkie components to be like ChonkieRecursiveChunker --> ChonkieRecursiveDocumentSplitter. We tend to use Splitter in our library rather than Chunker (fine to call it chunking in the docs) and we like to be explicit if the i/o type is a Document by putting it in the name of the component. If you could update the names of all components that would be great.

[yugborana]

@sjrl resolved all the issues.

[yugborana]

@sjrl now I think it looks fine

[yugborana]

@sjrl

[sjrl]

@yugborana one more minor comment then ready to go!

[yugborana]

@sjrl done

[sjrl]

Thanks for the contribution!