-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Lucas Skog edited this page Jul 10, 2026
·
3 revisions
Meeting notes that stay on your machine.
referat is a Windows desktop app that records your meetings (system audio + microphone), transcribes them and writes finished meeting minutes — a summary, decisions and action items. You decide where the AI runs: locally on your machine, on your company's internal server, or with a cloud provider. Nothing leaves your machine unless you choose it.
- Privacy by design. Audio, transcripts and minutes are stored locally. Outbound traffic goes only to endpoints you chose yourself: text to your AI services and — if you enable the optional speaker identification — audio to a local server you run. There is no telemetry.
- You choose where the AI runs. Point transcription and summarization at any OpenAI-compatible endpoint — a local server, your company's internal server, OpenAI or Azure OpenAI. Anthropic is also supported for the minutes.
- Simple. One button starts the recording; the app transcribes and summarizes automatically when you stop. No terminal, no jargon.
- Works with any meeting tool. referat records your computer's system audio, so it works with Teams, Zoom, Google Meet or an in-person meeting — no bot joins the call.
- Installer: latest release — Windows 10/11.
- Landing page: https://sockulags.github.io/referat/
- Installation — download, the SmartScreen note and the first-run walkthrough.
- Local AI Setup — run everything on your own machine with speaches and Ollama.
- Configuration — every setting explained: provider presets, base URLs, models, the minutes template and more.
- Speaker Diarization — optional "who said what" labels in the transcript, powered by a local companion server.
- FAQ — cost, storage location, supported services, privacy.
- Architecture — how referat is built, and its security hardening.
- Roadmap — what's planned next.
The app interface is currently Swedish, and the default minutes template produces Swedish minutes. The transcription itself follows the spoken language. An English UI is on the Roadmap. This wiki and the landing page are in English.