Clearminutes: local Whisper meeting transcription, no audio leaves your machine
I built Clearminutes after getting frustrated with cloud meeting tools. Otter, Fireflies, and even "bot-free" tools like Granola still send your audio to cloud APIs (Deepgram, AssemblyAI, OpenAI) for transcription. If your meetings contain anything confidential, that's not great.
Clearminutes runs Whisper.cpp, Parakeet, and Canary 1B v2 directly on your device - Metal/CoreML on Apple Silicon, CUDA on NVIDIA, Vulkan on AMD/Intel, CPU fallback. Your audio is processed locally. The only network call the app makes is to verify a Pro licence, and only if you activate one.
It does the full workflow: real-time transcription → AI summary → action item extraction → export.
For AI summaries you can use a local model (Gemma + many others, via our bundled inference engine, or Ollama with any model you want) or optionally connect your own OpenAI/Anthropic/Groq key — text transcript only, audio never leaves.
Runs on macOS (14.2+) and Windows 10+. Free tier has no minute caps or history limit, but there is no speaker diarization. Pro adds those plus integrations.
I'm curious: for those of you who've tried running Whisper locally, how do you think about the speed/accuracy tradeoff between small (150MB, fast) vs large-v3 (3GB, best)?
Replies