lazyflow

native macos dictation.

hold right ⌥, speak, release. lazyflow transcribes, applies the active app profile, and pastes at the cursor.

requirements

• macos 14+
• microphone and accessibility permissions
• an api key for the selected cloud provider

setup

1. open settings
2. add the api key
3. grant microphone and accessibility access

controls

input	action
hold right ⌥	record; release to transcribe and paste
menu bar mic	toggle recording
transcript edit	save correction data
⌘q	hide windows; keep lazyflow running
menu bar → quit	terminate

pipeline

audio → transcription → corrections → app profile → paste

app profiles define tone, formatting, protected terms, correction pairs, and custom instructions. minimal bypasses llm cleanup.

storage

• api keys: keychain
• transcripts and corrections: sqlite in ~/Library/Application Support/LazyFlow/
• profiles: user defaults
• telemetry: none
• cloud transcription sends audio to the selected groq, openai, or elevenlabs provider
• cloud cleanup sends text to the selected llm provider

release

1. increment MARKETING_VERSION and CURRENT_PROJECT_VERSION
2. archive with developer id, notarize, and staple LazyFlow.app
3. run scripts/publish-update.sh /path/to/LazyFlow.app v<version>

the script rejects invalid release signatures, generates the sparkle signature and appcast.xml, then publishes both assets to github. installed builds follow the latest appcast through sparkle.