Patter

Native macOS dictation app: speak into a capture session, watch a live transcript build in the app, and when you stop, the finished transcript is copied to your clipboard (⌘V it wherever you want) and always saved locally. See 1-pager.md for the full product spec.

Requirements

• macOS 26.2+, Xcode 26+
• xcodegen (brew install xcodegen) — this repo checks in project.yml, not the generated .xcodeproj

Installing as a standalone app

To build a Release version and install it to /Applications (no Xcode needed to run it afterward, and no manual setup step first — this generates the Xcode project for you if it doesn't already exist):

./scripts/install.sh

Re-run this after any code change to update the installed copy.

Developing in Xcode

xcodegen generate
open Patter.xcodeproj

Then build/run with ⌘R. On first run you'll be prompted for microphone access — that's the only permission this app needs; copying to the clipboard requires no special permission.

.envrc documents the Xcode toolchain path for command-line builds but requires direnv to take effect automatically — see the comment in that file.

Local model cache

WhisperKit downloads the whisper-large-v3-turbo CoreML model from HuggingFace on first use and caches it locally at ~/Documents/huggingface/models/argmaxinc/whisperkit-coreml/openai_whisper-large-v3-v20240930_turbo (~360MB); all transcription after that first download runs fully on-device with no network access. The app checks for this folder itself before ever calling out to HuggingFace, so later launches touch no network at all, not just "probably won't redownload much."