Contributing to Haystack

We welcome contributions from the community! Whether it is fixing a bug, adding a new feature, or improving documentation, your contribution is valued.

Getting Started

Development Setup

git clone https://github.com/deepset-ai/haystack.git
cd haystack
pip install -e ".[dev]"

Running Tests

# Unit tests
pytest tests/unit

# Integration tests (requires external services)
pytest tests/integration

# Specific component tests
pytest tests/unit/components/generators/

Contribution Areas

Components

Haystack's power comes from its component ecosystem. We welcome new components for:

Document converters and preprocessors
Embedding models and rankers
LLM integrations (new providers, fine-tuned models)
Document stores and retrievers
Custom evaluation metrics

Integrations

External integrations live in haystack-extras or as standalone packages:

New vector database backends
Cloud provider integrations (AWS, GCP, Azure)
Monitoring and observability tools

Documentation

Tutorial improvements and new tutorials
API documentation updates
Usage examples and best practices

Pull Request Process

Fork the repository and create a feature branch
Write tests for any new functionality
Format code with black and isort
Lint with pylint and type check with mypy
Document new components with docstrings
Submit a PR referencing any related issues

PR Checklist

Tests added or updated
Documentation updated
Changelog entry added
All CI checks pass
No breaking API changes (or discussed and approved)

Component Development Guide

Each Haystack component follows a standard interface:

from haystack import component

@component
class MyComponent:
    @component.output_types(output=str)
    def run(self, input_text: str) -> dict:
        return {"output": processed_text}

Components must:

Declare input and output types
Be serializable (for pipeline YAML export)
Include comprehensive docstrings
Have unit test coverage

Code of Conduct

We follow the Contributor Covenant code of conduct.

License

By contributing, you agree that your contributions will be licensed under the Apache License 2.0.