Horch

Horch is a privacy-focused meeting assistant for macOS that captures, transcribes, and organizes spoken conversations entirely on-device. Instead of joining calls as a bot the way Otter, Fireflies, or Read.ai do, Horch records at the operating-system level, then runs Whisper locally to produce transcripts without sending audio to the cloud. From those transcripts it extracts action items, generates topic summaries, and builds people pages that accrue discussion history and open follow-ups over time, so every recurring contact effectively gets a living dossier. Everything is written as plain Markdown into ~/Meetings, which means the resulting knowledge base is diff-able, greppable, and portable to any other tool that reads Markdown (Obsidian, Logseq, ripgrep, editor search). It also exposes those notes over the MCP protocol, so Claude, Cursor, and other MCP-aware assistants can query your meeting history directly and use it as long-lived context. A CLI is provided for scripting and batch operations. The intended user is a knowledge worker who runs many meetings, cares about confidentiality (legal, medical, executive, security work, or anyone under NDA), and doesn't want a SaaS vendor storing recordings or transcripts. Because processing is local, there's no per-minute transcription fee and no bot showing up in the participant list; because storage is Markdown on disk, the notes outlive the app itself. Optional local model backends (Ollama, MLX) let users choose their own summarization model without ever leaving the machine.