watzon/pindrop

Pindrop – Offline‑first AI dictation for macOS

What it is
Pindrop is a native macOS menu‑bar application that turns spoken words into text. It runs locally on Apple‑silicon Macs, using on‑device transcription models (e.g., Whisper‑Kit, Parakeet, SenseVoice, Apple’s own speech stack) and can fall back to cloud services like OpenAI’s GPT‑4o‑Transcribe when the user prefers. The app is written entirely in Swift/SwiftUI, stores data with SwiftData, and updates itself via Sparkle.

Key capabilities

• Multiple transcription back‑ends – Choose between several local models (Whisper‑Kit Core ML, Parakeet, SenseVoice, Apple Speech) or cloud‑based OpenAI GPT‑4o models.
• Streaming dictation – Words appear in real time, powered by the Nemotron streaming engine (English only) or Apple’s SpeechTranscriber on newer macOS releases.
• Multilingual support – 16 spoken languages, UI localized to 31 languages with right‑to‑left layout support.
• Speaker diarization – Offline diarization with trainable speaker profiles, useful for multi‑speaker recordings.
• Library & editing – Transcripts are saved automatically, searchable, editable in‑place, and exportable as JSON, CSV, or plain text.
• Notes & dictionary – Attach pins, tags, checklists, and a custom dictionary that learns from user corrections.
• AI‑enhanced cleanup – Optional post‑processing using Apple Intelligence, Anthropic, OpenRouter, Ollama, or any OpenAI‑compatible endpoint.
• Integration hooks – A built‑in MCP server lets other AI agents command Pindrop, and a context engine rewrites spoken file paths into @mentions.
• Privacy‑first defaults – All audio and transcription stay on the Mac unless the user explicitly enables a cloud model; API keys are stored in the macOS Keychain.

How to get it

• Download the signed DMG from the GitHub releases page.
• Drag the app to /Applications, grant microphone (and optionally Accessibility) permissions, and follow the onboarding wizard to download a model (the 75 MB “tiny” Whisper model is the default starter).

Typical workflow

1. Press the global hotkey (Option + Space by default) to start recording.
2. Speak – words appear live if streaming is enabled.
3. Press the hotkey again (or release in push‑to‑talk mode) to stop and generate the transcript.
4. The result is copied to the clipboard and, with Accessibility permission, inserted at the cursor in the active app.
5. Open the menu‑bar window to view, edit, tag, or export the transcript.

Why it matters
Pindrop showcases a fully offline, macOS‑native AI dictation pipeline that leverages Apple Silicon and Core ML for on‑device Whisper inference. It gives power users a privacy‑preserving alternative to cloud‑only dictation services while still offering optional cloud models and AI‑powered cleanup for those who need them.

Current status
The project is archived and officially discontinued (last version 1.23.0). The developer recommends migrating to Sayso, a successor app that can import Pindrop data. The source code remains MIT‑licensed and can be forked or studied, but no new contributions or bug fixes will be accepted.

Useful links

• Release page: https://github.com/watzon/pindrop/releases
• Documentation & build instructions are in the repo (BUILD.md, CONTRIBUTING.md).
• Replacement app Sayso: https://justsayso.app (source at https://github.com/watzon/sayso).