Installation¶
Install the package¶
The 0.3 release reports 0.3.0. The core supports Python 3.10–3.14 and does not
require torch or scikit-learn. Durable execution uses a local POSIX coordinator;
macOS and Linux are supported, while Windows and multi-host/network-filesystem
coordination are outside the 0.3 support boundary.
Install from source¶
git clone https://github.com/justinGrosvenor/alignmenter.git
cd alignmenter
python -m venv .venv
source .venv/bin/activate
pip install -e 'alignmenter[test,docs]'
The package directory is alignmenter/ inside the repository. Source installation
also gives you the repository's application fixtures and case studies.
Optional dependencies¶
Quote bracketed requirements so your shell does not expand them.
| Extra | Purpose |
|---|---|
[test] |
Core tests, Ruff, build, and distribution validation without ML extras |
[docs] |
MkDocs and Material documentation builds |
[ml] |
torch, sentence-transformers, and transformers for local embeddings/safety |
[calibrate] |
scikit-learn and numpy for numeric calibration |
[safety] |
Compatibility alias for [ml] |
[all] |
All optional runtime dependencies |
[dev] |
Existing full development extra, including ML/calibration dependencies |
ML extras have their own upstream platform requirements. Local classifiers and embeddings can download model weights on first use; they are not used by the offline resource-task example.
Provider credentials¶
The quickstart runs without credentials. Configure keys only for adapters that
actually call a provider, for example OPENAI_API_KEY or ANTHROPIC_API_KEY in the
process environment or your application's secret manager. Keep credentials out of
suite YAML, datasets, source control, and report artifacts.
Run the quickstart, or read the 0.3 migration guide for existing integrations.