Parrot — AI meeting recorder & live copilot for Mac, free & open | LaunchDaily
Parrot
AI meeting recorder & live copilot for Mac, free & open
meeting recorder
ai copilot
open source
on-device ai
transcription
speaker detection
Parrot is a free, open-source (GPL-3.0) AI meeting recorder and live copilot for Mac. It detects when a call starts, records the audio your Mac already plays and hears (no bot joins the meeting), transcribes it, and surfaces live copilot cards during the call — answers grounded in your own documents, with cited sources. After the call it produces a summary, action items and coaching with timestamps linked back to the transcript. All audio, transcription, speaker detection and document indexing can run entirely on-device; cloud AI is opt-in per model choice.
Key Features
No-bot recording
Records what your Mac already plays and hears, so nothing shows up in the participant list; works with Google Meet, Zoom, Teams or a phone on speaker, and asks whether to record when a call app opens.
Live copilot with document grounding
Indexes PDFs, text and markdown on-device using Apple's language models; matching excerpts appear within a second of a question, each card naming its source (your document or general knowledge), with a setting for whether it may answer beyond your files.
Call profiles
Seven built-in profiles (sales discovery, interview, 1:1 coaching, investor update, and more) or custom ones — each decides what the copilot watches for live (objections, buying signals, red flags) and how the post-call report is written; calendar events can auto-pick the right profile.
Speaker detection with names
On-device diarization with a 13 MB model tells voices apart, asks you who they were once, and can remember voiceprints so future calls suggest names — including one-click names from calendar invites.
Post-call report and coaching
Summary, action items, objections handled/missed, talk-balance gauge, and coaching notes — every point timestamped and clickable to hear the exact line; unverified promises are flagged so AI slips can't become commitments.
Bring your own AI
Run the copilot locally with Ollama, or supply keys for Claude, OpenAI, Gemini, Groq, OpenRouter, LM Studio or a custom server; every choice states exactly what leaves your Mac (e.g. transcript text, never audio).
Multilingual support
Auto-detects the call's language; Parakeet v3 covers 25 European languages free on-device, others fall back to Whisper; copilot, reports and document search work across languages (ask in English, find answers in Turkish or Spanish docs).
Claude and AI-app integration
Optional one-click read-only connection so Claude Desktop, Cursor, Claude Code or Codex can answer questions about your meetings, with weekly digests, follow-up drafts and call prep.
How It Works
1
Pre-call briefing
Type a line or two before hitting record so the copilot knows who you're talking to; standing coaching instructions apply to every call, and repeat meetings start with what was still open last time.
2
During the call
Live cards deliver document answers, objections, buying signals and pacing nudges; a pause button stops all sending and spending, and relaxed/fast pacing controls cost and latency.
3
After the call
A report written per the call's profile is ready before you hang up, with linked timestamps, commitments and follow-ups, plus coaching on what to improve.
Pros & Cons
Pros
Free and fully open source (GPL-3.0) with no account and no telemetry
Strong on-device privacy: audio never leaves the Mac, local models show $0.00, and on-device-only profiles keep everything local
No meeting bot joins your calls — it records what the Mac already hears
Per-call cost transparency: every meeting shows its real model/token cost
Bring-your-own-model flexibility, from fully local Ollama to Claude/OpenAI/Gemini keys
Cited, timestamped sources for both live cards and post-call reports
Who It's For
Best for
Mac users on Apple Silicon who want meeting capture without a bot joining the call
Salespeople, recruiters and consultants who want live, document-grounded answers mid-call
Privacy-conscious users who want transcription, diarization and AI running locally
Developers and tinkerers who want an open-source tool they can build from source and plug into their own models
Not ideal for
Windows or Linux users, or anyone on Intel Macs
Teams wanting cloud-native meeting infrastructure with centralized administration
Use Cases
Sales discovery calls with live objection handling and document-backed pricing answers
Job interviews with red-flag and strong-answer detection
1:1 coaching and investor updates via dedicated profiles
Searching and asking questions about past meetings from Claude, Cursor or Codex
Transcribing and summarizing existing audio recordings dropped into the app
Pricing
Parrot itself is free and open source (GPL-3.0), with no account required. On-device features (recording, transcription with Whisper/Parakeet, speaker detection, Ollama copilot) cost nothing. If you use a cloud copilot with your own API key, you pay the provider directly — the site shows an example of roughly $0.07 for one hour of calls using Claude ($0.06 live cards + $0.01 post-call report, $0.0
Integrations
Google Meet
Zoom
Microsoft Teams
Ollama
Claude
OpenAI
Gemini
Groq
OpenRouter
LM Studio
Custom server
Whisper
Parakeet
Deepgram
Claude Desktop
Claude Code
FAQ
Is Parrot really free?
Yes. Parrot is free and open source under GPL-3.0, with no account and no telemetry. On-device features cost nothing; if you use cloud AI like Claude, you pay that provider with your own API key (the app shows the exact cost per meeting, e.g. roughly $0.07 per hour of calls in its example).
Does a bot join my meetings?
No. Parrot records what your Mac already plays and hears, so nothing appears in the participant list. It works with Google Meet, Zoom, Teams, a phone on speaker — anything your Mac can hear — and asks whether to record when a call app starts.
Does my audio leave my Mac?
No. Recording, transcription, speaker detection and document indexing all run on your machine. Cloud AI is opt-in: only the transcript text is sent to the chosen provider (never the audio), and every model choice states exactly what it sends.
macOS-only — requires macOS 14 or later on Apple Silicon, so Windows and Linux users (and Intel Macs) are out of luck
Cloud copilot quality (e.g. Claude) requires your own API key, and transcript text goes to the provider's servers
Local-only operation depends on Ollama or local Whisper/Parakeet, which may be slower or lower quality than cloud models
Language coverage for the free on-device Parakeet model is limited to 25 European languages (excluding Turkish), with other languages needing Whisper
The AI-app integration sends queried content to Anthropic/OpenAI/Cursor under your account, a different privacy mode from fully on-device calls
Users who won't supply any API key and need top-tier cloud model quality rather than local models
Anyone needing voice-tone analysis — the live nudges read timing and words only, never tone
Parrot runs on macOS 14 or later on Apple Silicon. Version 0.25.0 is a 23 MB signed and notarized download, and you can also build from source.
Which AI models can I use?
Run everything locally with Ollama, or bring API keys for Claude, OpenAI, Gemini, Groq, OpenRouter, LM Studio or a custom server. For transcription, choose Whisper or Parakeet on your Mac, or cloud options Groq and Deepgram. "Ask Parrot" can also pick its own model per task.
What languages does it support?
Language is auto-detected per call. Parakeet v3 handles 25 European languages for free on-device (no Turkish); other languages use Whisper. The copilot, reports and document search are multilingual — an English question can find the answer in a Turkish or Spanish document.
How does the copilot use my documents?
Drop in PDFs, text or markdown; they're chunked and embedded on your Mac with Apple's language models and never uploaded. Live cards quote them with a named source, and you control whether the copilot may answer beyond your files at all.
What are profiles?
Profiles tell Parrot what kind of call it is — sales discovery, interview, 1:1 coaching, investor update and more (seven built-ins, or custom). Each profile changes what the copilot watches for live and how the post-call report is written, and calendar events can auto-select the right one.
