EchoDiar

Record any meeting on your Mac or Windows PC, see who said what, and ask AI about it — local-first and lightweight.

macOS 15+ · Apple Silicon · Windows 11 (x64, Arm64) · Free & open source (AGPLv3)

Installed EchoDiar on Windows with its certificate before? Move to the Microsoft Store; your recordings stay.

Features

  • Lightweight and native: a Swift app in the Mac menu bar, a WinUI app in the Windows notification area. Models load only while they work.
  • Private: transcription and AI can run fully on your computer. Audio goes to the cloud only if you choose a cloud engine.
  • Any meeting app, any AI: record Zoom, Teams, Meet in a browser or all system audio. Ask local models, cloud APIs or your Claude Code or Codex subscription.
  • Who said what: your mic and the meeting audio on separate channels, the other speakers told apart and renamable.
  • On-device transcription: Whisper and Parakeet on Mac and Windows, plus Qwen3-ASR and Apple Speech (macOS 26) on Mac. Whisper and Parakeet transcribe while you record, with live captions, so the transcript is ready seconds after you stop.
  • AI summaries and chat: answers cite timestamps that jump to that moment in the recording.
  • Copy as prompt: the transcript and your question, ready to paste into any chat app.
  • Meeting detection: offers to record when a call starts, shows a floating indicator and offers to stop when the meeting ends.
  • Library and search: conversations by day, search over titles and transcripts, auto-delete of old recordings.

Transcription

  • On your computer: Whisper (WhisperKit on Mac, whisper.cpp on Windows) and Parakeet, plus Qwen3-ASR and Apple Speech (macOS 26) on Mac.
  • Speakers: diarization of the meeting audio (FluidAudio on Mac, sherpa-onnx on Windows), when "Tell speakers apart" is on (off by default).
  • Transcribe while recording: Whisper or Parakeet transcribes each channel as you speak, and the live captions come from it. On a Mac, for a 15-minute call, the transcript is ready 0.8 seconds after Stop (about 10 seconds with speakers told apart) instead of about a minute and a half. Other engines transcribe the whole recording after Stop.
  • Cloud (optional): OpenAI, Groq, Mistral, Gemini, xAI, ElevenLabs, Deepgram, AssemblyAI, Speechmatics, Soniox, Gladia or an OpenAI-compatible endpoint, with keys in Settings > AI Providers > Cloud. Asks once before audio goes to each one. A live transcript can then come from a local model (free) or the provider's realtime stream (paid, on top of the transcription after Stop).

AI providers

  • On-device: MLX models and Apple Foundation Models (macOS 26) on Mac; GGUF models, and Phi Silica on Copilot+ PCs, on Windows.
  • Local servers: LM Studio, Ollama, Jan, llama.cpp and vLLM, plus Foundry Local on Windows.
  • Cloud APIs: OpenAI, Anthropic, xAI, Gemini, Mistral, Azure OpenAI, OpenRouter, Groq, DeepSeek or a custom endpoint, with your own key in Settings > AI Providers > Cloud.
  • Subscription CLIs: Claude Code, Codex, OpenCode and Gemini CLI.

Get started

  1. On Mac, download the signed and notarized DMG from the releases page and drag EchoDiar to Applications. On Windows 11, get EchoDiar from the Microsoft Store.
  2. Allow the microphone (and System Audio Recording on Mac), then download a model in Settings > AI Providers > Local AI.
  3. Start and stop recording from any app with Cmd+Option+R on Mac or Ctrl+Alt+R on Windows. The transcript follows.

More in the docs. EchoDiar updates itself.