Your meetings, your silicon: a local AI notetaker

What it does

Meetily is a desktop meeting assistant that records audio, transcribes speech in real time using Whisper or Parakeet models, and generates AI summaries — all without network egress. It runs as a Tauri app with a Rust backend and Next.js frontend, targeting macOS and Windows with GPU acceleration via Metal, CUDA, or Vulkan.

The interesting bit

The project treats cloud dependency as a bug, not a feature. It bundles local LLM support through Ollama and also permits custom OpenAI-compatible endpoints, letting organizations with existing AI infrastructure plug in without rerouting data. The “Import & Enhance” feature can re-process existing recordings with different models or languages, which is useful when your first-pass transcription turns out to be gibberish.

Key highlights

• Real-time transcription with speaker diarization using local Whisper/Parakeet models
• GPU acceleration auto-enabled at build time for Apple Silicon, NVIDIA, and AMD/Intel GPUs
• Summary generation via Ollama (recommended), Claude, Groq, OpenRouter, or custom endpoints
• Simultaneous microphone and system audio capture with ducking and clipping prevention
• MIT-licensed community edition; separate “Meetily PRO” and “Enterprise” tiers exist with different codebases

Caveats

• Linux users must build from source; no prebuilt binaries are provided
• Several PRO-tier features — speaker identification, calendar integration, meeting chat — are marked “Coming Soon” even in the paid product
• The “4x faster” transcription claim in the repository description is not substantiated with benchmarks in the README

Verdict

Worth evaluating if your compliance team breaks out in hives at the words “SaaS notetaker.” Skip if you need polished mobile support or guaranteed speaker diarization today; the community edition’s diarization capabilities are not demonstrated in the provided documentation.