OpenDecision v0.2.0

OpenDecision 0.2.0 adds image decisions with google/siglip2-base-patch16-224.

• Send an image and typed questions to /v1/systemone. Choice, Noul, Score, and Relation keep their existing answer shapes.
• Install image support with pip install 'OpenDecision[images]'. The image engine loads on the first valid image request.
• Combine answers into alarm, collection, and repair decisions using the existing rule evaluator.
• View the bear, overflowing-bin, and damaged-fence examples in the README and documentation, with measured local HTTP call times.
• Run editable image cases, generate a visual report, and reproduce the API timings from the included demo scripts.

Validation: 73 tests passed. Wheel and source archive passed Twine checks. The documentation built in strict mode, and the wheel installed in a clean environment.

Image scores are uncalibrated. The examples document counting, negation, and material-classification errors. Timings were measured on an Apple M3 with preloaded models.

OpenDecision v0.1.2

OpenDecision 0.1.2 keeps ModernBERT as the default model and documents DeBERTa v3 large as an explicit option for Python and server users.

The release also caps direct NLI inputs to the selected model's context window, records the model comparison, and reports the configured model in benchmark output. The standard Choice behavior remains unchanged.

Validation: 55 tests passed; the documentation and package distributions built successfully.

OpenDecision v0.1.1

OpenDecision 0.1.1 is the first PyPI package release.\n\nHighlights:\n- Install with pip install OpenDecision or uv add OpenDecision.\n- Import OpenDecisionEngine from the top-level opendecision package.\n- Start the API with opendecision serve.\n- Answer questions over documents with binary, three-way, or combined NLI.\n- Retrieve source evidence and compose decisions with auditable rules.\n- Run the real-time ViZDoom demonstration.\n- Publish through PyPI Trusted Publishing.