Pawscribe Changelog

Release notes for Pawscribe (声痕).

v1.0.3 — Pick any AI model

Model dropdown for every provider. AWS Bedrock, Anthropic, OpenAI, Gemini, DeepSeek, Qwen, GLM, Kimi, OpenRouter, Ollama — each now offers today's mainstream models (Claude Fable 5 / Opus 5 / Sonnet 5, GPT-5.6/5.5, Gemini 3.6, DeepSeek V4, GLM-5.2, Kimi K3…).
Free-form model names. The model field is editable, so a brand-new model works the day it launches.
Self-updating list. The model list refreshes from the server every 24h (plus a manual refresh button) — no app update needed when new models ship.
Per-provider memory. Each provider remembers its own model; switching providers no longer overwrites your choice.

One-click verification. A new Test Connection button sends a minimal request with your Base URL, API key and model, confirming the setup instantly — with a clear error message when something's off.

Select and copy across the whole AI reply. Drag-select across headings, paragraphs and lists and copy the entire span at once (rich text) — no more copying one block at a time.

Clickable timestamps. Action-item timestamps now show the in-recording position first — 07:51 (16:17:04) — and clicking jumps to that transcript line and starts playback.

Import now just brings the file into your library — no cloud key required up front. Afterwards you choose how to transcribe from the recording's Transcribe menu: AWS high-accuracy (online) or local Whisper (offline). Video files can be transcribed offline too.

The AI chat panel can now be dragged wider to the left.
Fixed: the recordings list column could be dragged shut and lost for good. It now has a minimum width, plus a show/hide button.

Fixed a crash when running local Whisper on Intel Macs. WhisperKit on Apple Silicon (ANE-accelerated), whisper.cpp on Intel (CPU) — same UI, the app picks the right engine automatically. One Universal download runs natively on both.

First public release. Everything you need to record, transcribe, and make sense of your meetings.

Dual-track recording. System audio and your microphone saved as two separate tracks.
Auto meeting detection. Zoom, Teams, Google Meet, FaceTime, Tencent Meeting, Feishu — Pawscribe notices when a meeting starts and prompts you with one click to record.
Echo cancellation. WebRTC AEC3 keeps your mic clean even when you're on speakers.
Quiet menu-bar app. No Dock icon, no bouncing, no broadcast during screen shares.

Real-time bilingual captions. Live subtitles for both tracks, each independently set to its own language.
Apple on-device speech recognition. Fast, local, no cloud needed.
Optional Whisper transcription. Local OpenAI Whisper for higher-accuracy re-transcription. Translate any meeting to English in one click.
AWS Transcribe support for languages where you want cloud-grade accuracy.
Speaker diarization. Local FluidAudio CoreML separates speakers on the remote track. Name them once — the names follow into transcripts and exports.

AI meeting summaries. Summary, decisions, action items, and a follow-up message — drafted the moment your meeting ends.
12 AI providers. AWS Bedrock, Anthropic Claude, OpenAI, Gemini, Azure OpenAI, DeepSeek, OpenRouter, Ollama, Alibaba Bailian (Qwen), Zhipu GLM, Moonshot Kimi, plus on-device Apple Foundation Models. Bring your own key.
Meeting Brain. Floating window with ⌘⇧A. Ask anything across your entire meeting library — answers come with citations that jump back to the original audio.
Cross-meeting action items. Every to-do extracted from every meeting, in one panel.

Built-in player with subtitle sync. Click a line to jump to that moment. Edit subtitles in place.
History, search, and projects. Full-text search across transcripts and notes. Organize recordings into projects with color tags and drag-and-drop.
Markdown export. Transcript + notes + AI summary in one clean .md file.

"Never leave local" switch. One toggle disables every cloud path at once.
Standard files. Recordings are plain .m4a in ~/Documents/Pawscribe. No proprietary format, no accounts, no telemetry.
Bring your own keys. Cloud providers are opt-in. Credentials stored in macOS Keychain.