Every AI notetaker on the market asks you to make the same trade: convenience in exchange for a copy of your most sensitive conversations. Your product roadmap, your legal strategy, your patient consult, your salary negotiation — all of it uploaded, transcribed, and stored on a server you will never see. Meetily rejects that bargain entirely. It is an open-source, privacy-first meeting assistant that captures, transcribes, and summarizes your meetings without a single byte leaving your machine.
With 27,400 GitHub stars and a permissive MIT license, Meetily has become the most popular self-hosted meeting notetaker for macOS and Windows. And unlike the polished SaaS incumbents, it earns that popularity by doing the one thing they structurally cannot: keeping your data yours.
The privacy problem nobody wants to name
Cloud meeting tools are convenient right up until they aren't. The Meetily project frames the risk in blunt numbers, and the underlying figures — if anything — undersell it. IBM's Cost of a Data Breach report put the global average breach at $4.88 million in its 2024 edition, a 10% jump over the prior year. Cumulative GDPR enforcement has since climbed past €7 billion in fines. For a defense consultant, a healthcare provider, or an in-house counsel, "we sent your meeting audio to a third-party API" isn't a footnote — it's a compliance incident waiting to happen.
Whether you're a legal professional, an enterprise executive, or a founder discussing an unannounced acquisition, your sensitive discussions shouldn't live on servers you don't control.
The uncomfortable truth is that most AI notetakers are built cloud-first because it's cheaper for the vendor. Meetily inverts the incentive: it uses open-source models running on your own hardware, so there's no API bill and no data exfiltration in the first place.
What Meetily actually does
At its core, Meetily is a desktop app built with Tauri, pairing a Rust backend for the heavy lifting with a Next.js frontend for the interface. That architecture matters — Rust gives it the performance to run speech models in real time without the memory bloat of an Electron wrapper.
The feature set is genuinely complete for daily use:
| Capability | How it works |
|---|---|
| Local transcription | Runs Whisper or NVIDIA's Parakeet models entirely on-device |
| Speaker diarization | Separates who-said-what using a sortformer-based pipeline |
| AI summaries | Generates minutes via Ollama (local), or Claude, Groq, OpenRouter, or any OpenAI-compatible endpoint |
| Dual audio capture | Records mic and system audio with intelligent ducking to prevent clipping |
| Import & enhance | Bring in existing recordings and re-transcribe with a different model or language |
The flexibility on the summarization layer is the smart part. Purists can keep the entire pipeline offline with Ollama. Teams that already trust a provider can point Meetily at Claude or Groq for higher-quality summaries while still keeping transcription — the part that touches raw audio — completely local.
GPU acceleration that just works
Speech-to-text is compute-hungry, and Meetily leans on whatever silicon you have. On Apple Silicon it uses Metal and CoreML; on Windows and Linux it targets NVIDIA CUDA and AMD/Intel via Vulkan. Crucially, this is wired up at build time — there's no config file to hand-tune. The project claims roughly 4× faster transcription with the Parakeet path versus baseline Whisper, which tracks with Parakeet's reputation as a fast, accurate ASR model.
Installing Meetily
On macOS and Windows, installation is a normal desktop download — no terminal required.
- macOS: grab
meetily_0.4.0_aarch64.dmgfrom the latest release, open it, and drag Meetily to Applications. - Windows: download the
x64-setup.exeinstaller and run it.
Linux users build from source, which is also the path for anyone who wants to audit or modify the app:
git clone https://github.com/Zackriya-Solutions/meeting-minutes
cd meeting-minutes/frontend
pnpm install
./build-gpu.sh
You'll need Rust and Node.js installed first. The build-gpu.sh script detects your hardware and compiles the accelerated transcription backend automatically.
The honest caveats
Meetily is not pretending to be a finished consumer product, and you shouldn't treat it as one. The macOS build ships as a 0.4.0 release — this is pre-1.0 software, and the issue tracker (200+ open issues) reflects the churn you'd expect from a fast-moving project. Import-and-enhance is explicitly labeled Beta.
There's also a commercial angle to understand. The maintainers, Zackriya Solutions, sell a Meetily PRO tier built on a separate codebase with higher-accuracy models, advanced exports, and auto-join. That's a legitimate open-core business model, but it means some of the most polished features — custom summary templates, DOCX/PDF export, calendar integration — sit behind a paywall. The Community Edition is promised to stay free and open source, and for local transcription plus AI summaries, it already delivers.
The Bottom Line
Meetily is the clearest answer yet to a question more teams are finally asking: why does an AI notetaker need to see the cloud at all? It doesn't. With local Whisper and Parakeet transcription, flexible summarization that can stay fully offline, and an MIT license that lets you self-host and audit every line, Meetily hands data sovereignty back to the person in the meeting. It's rough in the way pre-1.0 software always is, and the PRO upsell is real. But if your conversations are the kind you'd never paste into a public chatbot, a 27K-star tool that keeps them on your own hard drive isn't just appealing — it's the only sane default.
