Parrot

Parrot is a free, open-source (GPL-3.0) meeting recorder for Mac, made by Uygar Turantekin, with a built-in live AI assistant. It records and transcribes calls on the machine, suggests answers from the user's own documents while a call is in progress, and writes a report afterward. The current release listed on the site is 0.29.7, for macOS 14 or later on Apple Silicon.

What It Is

Parrot is a native SwiftUI desktop app that captures system audio and the microphone as two tracks, so no bot joins the meeting. It works with Google Meet, Zoom, Teams, or anything else the Mac can hear. Transcription, speaker detection and the document index run on the Mac; the developer states there is no account, no telemetry and no Parrot server.

How a Call Works

Before a call, users can add a short brief and drop PDFs, text or Markdown files into a knowledge base. These are chunked and embedded on the Mac using Apple's NaturalLanguage framework. During the call, the Assistant shows suggested answers that name their source, pinned cards for objections and open questions, captured next steps, a live call score and nudges when the call drifts. Call profiles (eight built-ins, or custom ones) decide which card types and gauges appear and how the post-call report is written.

After the Call

The report includes a summary, talk balance, objections marked handled or missed, coaching, and commitments, with time chips that link each point to the transcript line. Speaker detection runs on-device after the call and lets users name voices. Reports export as PDF, and notes can go to Obsidian, Reminders, follow-up email drafts or a webhook. Ask Parrot (⌘K) answers questions about past calls, and Claude Desktop, Claude Code, Codex and Cursor can read meetings through an opt-in connection. Existing audio files can be imported.

Model Choices and Privacy

Transcription can use on-device Whisper or Parakeet, or Groq or Deepgram with the user's own key. The Assistant can run on Claude, a local model through Ollama, or any OpenAI-compatible server. Running Whisper and Ollama keeps everything on the Mac and works offline after the models download. With cloud options, transcript text goes to the provider the user picked, and audio leaves only if Groq or Deepgram is chosen for transcription.