What's new
- Slide deck generation — turn any note into a structured slide deck via Claude / OpenAI / Gemini, export as Markdown
- Podcast generation via ElevenLabs — turn a note into narrated audio (set ElevenLabs API key in Settings)
- Prompt caching for Claude — ~90% lower cost on multi-turn chats with stable system prompts
- RAG citations grouped by document with multi-passage navigation in the source modal
Fixes
- Chat returning empty results when stale vectors lingered in the HNSW index — search now filters to chunks that still exist in SQLite
- Opening the app no longer creates a new "Untitled Note" every time — opens your most recent existing note instead
- Conversation naming now works with GPT-5.x models (was failing on
max_tokensrejection) - Quiet speech now reaches Whisper (removed the aggressive RMS gate that dropped soft / distant audio)
- Build script clears stale DMG artifacts before rebuilding (avoids
bundle_dmg.shfailing silently on leftover files) - Local Whisper log spam suppressed; Windows compile fixed via
c_intlog callback
Install
Download the .dmg below, open it, drag Platypus to Applications. Signed and notarized.
- macOS Apple Silicon only (Intel Mac and Windows builds still in progress)
- First local transcription run will download the Whisper model (~3.1GB for Large v3, the new default)
Build from source
See README. Now includes Windows-specific instructions for cmake + LLVM/libclang.