v.1.0.1

1.0.1 - 2026-08-07

🐛 Bug Fixes & Improvements

• Fixed the --help description of gram check to clarify that it also checks database completeness by default, unless --skip-db is used.
• Reduced the AI provider costs of running gram import:
  • Prompt caching is now enabled, so retries reuse the cached prompt instead of paying full price.
  • The command now sends a minimized recipe payload instead of the full raw webpage data.
  • The self-correction loop no longer spends retries on informational warnings (like missing database ingredients).
  • Formatting is now handled locally instead of spending an AI call to fix it.
• Improved the quality and correctness of recipes generated by gram import:
  • Fixed an issue where English words would leak into non-English recipes.
  • Fixed grammatical errors (dropped articles) around ingredient and cookware references in non-English output.
  • Descriptive adjectives (like "boneless skinless") are now correctly extracted as preparation steps instead of creating unwieldy ingredient names.
  • Corrected the generated syntax for multi-word cookware references.
  • Fixed an issue where the target language was sometimes incorrectly inferred.
  • Improved decoding of accented characters and punctuation from the source recipe.
• Improved the CLI guidance printed after running gram import:
  • It now correctly points to gram db sync as the recommended next step.
  • It accurately reports any remaining compiler warnings (e.g., missing ingredients) instead of silently discarding them.
• Fixed gram init to ensure the interactive AI provider setup accurately offers the currently recommended models (e.g., llama4).