RecMeets — private meeting recording with local transcription

Coming soon. RecMeets is in the works and will be released soon. This page tracks its progress while it is being built — what is done, what is left, and what to expect from the first release.

The idea is simple: a Teams, Meet or Zoom meeting gets recorded and transcribed on your own computer. No bot joins the call. No audio file leaves the machine. At the end you get a transcript with speaker names, chapters and a player that jumps to the moment you need.

What it is

RecMeets is an app for Linux (any distribution with PipeWire or PulseAudio) and Windows 10/11. It records two separate tracks: your microphone and the computer's audio — you and everyone else. Speaker detection starts from a clean split, and the transcript knows which side is talking.

Speech recognition runs locally with whisper.cpp, accelerated on the GPU through Vulkan with a CPU fallback. Speaker detection runs locally too, through ONNX Runtime.

What it will do

• Two-track recording with pause and resume.
• Local transcription — the audio is never uploaded.
• Speaker detection, with renaming and fixes inside the app.
• Chapters for long meetings, optionally written by a local AI helper.
• Player and transcript editor: edits, speaker swaps, undo.
• File import (mp3, m4a and more) to transcribe older recordings.
• Crash recovery — an interruption does not lose the meeting.
• Post-meeting actions: your own scripts that run when a recording ends.
• Command line to start, stop and transcribe files.

Progress so far, phase by phase

The work is split into phases, each closed by a gate. So far:

• Foundations — done. New name and identity, icons, documentation and credit to the original project. All tests pass, Vulkan acceleration works, and a first command-line transcription came out right, with two speakers.
• Risk checks — in progress. Recording the microphone and system audio together on Linux passed with no timing drift. The Windows build and system-audio capture on Windows are being checked now.
• Platform layer — started. Audio encoding inside the app, with no external ffmpeg. New recording and playback shared by Linux and Windows. The player and file import already run on the new code.
• Next. More recognition engines — including a stronger one for Greek — separate languages for your side and the remote side, and GPU selection.
• New interface. A recording page, a compact control strip, a tray icon and multiple windows.
• Packages and release. AppImage and tar.gz for Linux, an installer and a portable build for Windows. After that, the first release.

Technologies used

• Rust — a single executable, fast and free of heavy dependencies.
• whisper.cpp through whisper-rs — local speech recognition on Vulkan or CPU.
• ONNX Runtime — local speaker detection.
• cpal — audio recording and playback on Linux and Windows.
• Opus — compressed recordings without external tools.
• Slint — the new interface on the way.

Origins

RecMeets started as a port of Meeting Recorder by Jankees van Woezik (MIT licence), which ran only on Omarchy/Arch. Here it becomes a standalone app for any Linux and for Windows.

When

Soon. The first release ships once the phases above are closed and have passed their gates. Until then, this page will be updated as the work moves on.