create-heed

One command to install heed — local-first meeting transcription with real speaker diarization. Built for Apple Silicon, running on Linux too.

npx create-heed

Your audio never leaves your machine. Ever. No API keys, no cloud, no subscription.

Why heed

Every other meeting tool ships your conversations to someone else's servers. heed doesn't. It transcribes, separates every voice, and writes AI notes entirely on your computer — so your work calls (NDAs, client data, strategy) and your private ones stay yours.

• 100% local & private — transcription, diarization, and AI notes all run on-device. Nothing is uploaded.
• Real-time and post-stop — live captions as people talk, plus a precise re-pass with real timestamps when you stop.
• Real speaker diarization — tells people apart by how they sound, not by login name — even on a single mic.
• Voice memory — name a speaker once; heed recognizes their voice in every future meeting.
• Overlap detection — catches two people talking at once instead of blending them into one line.
• Floating panel — an always-on-top window over Zoom, Meet, Teams, or Discord.
• AI notes — summaries and action items via a local LLM (Ollama), sized to your hardware.

The engine

• macOS (Apple Silicon) — Parakeet ASR + FluidAudio diarization run on the Apple Neural Engine via CoreML. No CUDA, no gated tokens, ~50× real-time. This is heed's fastest, most accurate path.
• Linux — faster-whisper + pyannote on CUDA (or CPU). Supported and actively evolving.
• Windows — coming.

Install

npx create-heed

The installer detects your machine and sets everything up: Bun, a supported Python, ffmpeg, the on-device AI engine, and (optionally) Ollama for AI notes. One command, no config. Then heed opens on http://localhost:5170.

Update

npx create-heed update

Pulls the latest, syncs dependencies, and rebuilds the engine only if it changed.

Troubleshooting

npx create-heed doctor     # check your install end-to-end
npx create-heed fallback   # install the fallback engine if the fast one isn't available

Local. Open. Yours. · MIT · Inspired by trx from CrafterStation

GitHub · Report an issue