A framework-agnostic Python library to seamlessly integrate Apache Cassandra with ML/LLM/genAI workloads.

Note: this is currently an alpha release.

Installation is as simple as:

pip install cassio

For example usages and integration with higher-level LLM frameworks such as LangChain, please visit cassio.org.

To develop cassio, use the requirements-dev.txt.

To use the dev version in an integration (e.g. your branch of LangChain), pip install -e . in this cassio repo from within the virtual environment you are using to develop your integration.

If the integration is Poetry-based (e.g. LangChain itself), you should get this in your pyproject.toml:

cassio = {path = "../../cassio", develop = true}

Then you do

poetry remove cassio                                      # if necessary
poetry lock --no-update
poetry install -E all --with dev --with test_integration  # or similar, this is for langchain

Inspired from this. You also need a recent Poetry for this to work.

Note: this is a rough guide, use your common sense and consider exceptions.

main is where versioned releases occur (the corresponding commits are git-tagged as v0.0.7 and similar, parallel to the PyPI releases).

Please do not push/PR more than trivial fixes directly to main.

Features should generally be done on specific branches and merged into dev for wider testing. Then, with a slower cadence (such as, when releasing), dev would be merged to main.

We are still at 0.*. Occasional breaking changes are to be expected, but please think carefully. Later, a stronger versioning model will be adopted.

Style is enforced through black and linting with ruff. We also take type-checking seriously. The code should run through make format without issues.

At the moment we try to run tests under Python3.8 and Python3.10 to try and catch versions-specific issues (such as the newer typing syntax such as typeA | typeB, illegal on 3.8).

• Bump version in setup.py
• Add to CHANGES.txt
• Commit the very code that will be built:

make build
twine upload dist/*  # (login to PyPI ...)

Please run tests (and add some coverage for new features). This is not enforced other than to your conscience. Type make for the available tests.

To run the full tests, there's make test-all.

You need a virtualenv with the requirements-dev.txt installed.

make test-unit

You need a virtualenv with the requirements-dev.txt installed.

Create the DB connection settings file, cp TEMPLATE.testing.env .testing.env and then edit the properties. You need at least one of either Astra DB or a Cassandra cluster to use, with vector-search support.

Launch the tests with either of:

make test-integration
make test-astra-integration
make test-cassandra-integration

The latter two above specify TEST_DB_MODE as either LOCAL_CASSANDRA or ASTRA_DB. Ideally you should test with both, since some tests are skipped in either case.