v—New
Model-agnostic builds by default
- FeatureEvery engagement now ships with a model-swappable adapter layer — move between OpenAI, Claude and Gemini without touching application code.
- FeatureStandard evaluation harness added to the delivery template: a task-specific test set and scoring pipeline as part of every build.
- FixTightened re-ranking defaults in the retrieval starter to cut near-duplicate passages in grounded answers.