# meetily

Privacy-first AI meeting assistant that runs the whole pipeline locally — captures mic and system audio, transcribes in real time with Whisper or Parakeet, and generates summaries via Ollama (or, if you opt in, Claude / Groq / OpenRouter / a self-hosted OpenAI-compatible endpoint). Single self-contained Tauri app: Rust backend, Next.js frontend, ships as a desktop binary on macOS and Windows, builds from source on Linux. MIT-licensed Community Edition with a separate paid PRO product on a different codebase for the speaker-identification / templates / DOCX-export tier.

The default-local-everything posture is the differentiator from Otter.ai / Fireflies / Granola / etc. — recordings, transcripts, and the transcription model itself stay on disk. Cloud LLM providers are an optional choice for the summary step only, and even that can be replaced with a local Ollama model.

## Architecture

- **App shell**: [Tauri](https://tauri.app/) — single-process Rust binary with a webview frontend
- **Frontend**: Next.js + TypeScript
- **Transcription**: Whisper (via [whisper.cpp](https://github.com/ggerganov/whisper.cpp)) or NVIDIA's Parakeet (via [transcribe-rs](https://crates.io/crates/transcribe-rs); the README credits an ONNX conversion of Parakeet-TDT-0.6B-v3 by `istupakov`)
- **Audio capture**: simultaneous mic + system audio with ducking and clipping prevention; some code borrowed from [Screenpipe](https://github.com/mediar-ai/screenpipe)
- **GPU acceleration**: Metal + CoreML on Apple Silicon, CUDA on NVIDIA, Vulkan on AMD/Intel — wired up at build time, no user config
- **Summaries**: Ollama (recommended), Claude API, Groq, OpenRouter, or any custom OpenAI-compatible endpoint

## Install

macOS aarch64 and Windows x64 ship as installers from [GitHub Releases](https://github.com/Zackriya-Solutions/meeting-minutes/releases/latest). Note the release artifacts live under the **`meeting-minutes`** repo name (the original name; the current canonical repo is `meetily`).

Linux is build-from-source per the [building_in_linux.md](https://github.com/Zackriya-Solutions/meetily/blob/main/docs/building_in_linux.md) doc:

```
git clone https://github.com/Zackriya-Solutions/meeting-minutes
cd meeting-minutes/frontend
pnpm install
./build-gpu.sh
```

## Import & enhance

Beta feature contributed by Jeremi Joslin et al — load an existing audio file to generate a transcript, or re-transcribe a previously recorded meeting with a different model or language. Same local-only pipeline.

## Community Edition vs PRO

The README is unusually explicit about the split. The OSS Community Edition is "forever free & open source" with local transcription, AI summaries, and core features. **PRO is a separate codebase**, sold on the meetily.ai website, with custom summary templates, PDF/DOCX exports, auto-detect-and-join, calendar integration, and built-in GDPR audit trails. Speaker diarization and chat-with-meetings are listed as "coming soon" in PRO. Enterprise is a third tier for 100+ users.

The CE/PRO split is worth noting because it shapes what improvements land where — the high-margin features (templates, advanced export, speaker ID, calendar, compliance) are deliberately reserved for the proprietary fork. The CE stays at the "transcribe + summarize" level.

## Comparable tools in the vault

- [[ghost-pepper]] — local STT for macOS, but for hold-to-talk dictation rather than meeting capture; same Whisper-based / privacy-first / Apple-Silicon-only pattern at smaller scope
- [[titit-local-ai]] — the broader case for local open-source LLMs over cloud APIs; meetily is a worked example of that argument applied to meeting assistants
- [[pion-handoff]] — the other place to tap a call: instead of recording what the OS plays back, it takes over the browser's WebRTC session and receives the media server-side as raw RTP, which sidesteps system-audio permissions entirely but only works where you control the signaling path

## Repo

[github.com/Zackriya-Solutions/meetily](https://github.com/Zackriya-Solutions/meetily) — 11.5k stars, MIT, Rust primary, repo created 2024-12-26, last push 2026-03-16. Active.