v0.1.5 (beta)
Major quality fixes for transcripts and summaries
- Punctuation and capitalization work again — a model-behavior change in macOS 26.5 made the on-device rewrite silently return text unchanged, so transcripts came out as raw lowercase. The rewrite path was reworked and now punctuates reliably (0 → 23 of 30 utterances in our benchmark), with a clean fallback when the model declines.
- Long meetings summarize fully — summaries no longer drop later sections of long transcripts, and action items consistently keep their owners. Behind the scenes, summarization is now budgeted by the model's actual token window instead of word-count guesses, so it also scales automatically to future macOS releases with larger windows.
Full Changelog: v0.1.4...v0.1.5