Definition
Leclog is a local-first desktop application for lecture sessions. It is built as a Tauri 2 app with a React and TypeScript frontend, a Rust command backend, and JSON-file session persistence.
The app is designed to record lecture audio, generate local transcripts, and keep the resulting session files, models, and processing artifacts under user-managed local storage.
Current scope
- Create and browse lecture sessions.
- Record, pause, resume, and stop audio capture.
- Capture local microphone audio or macOS system audio.
- Generate transcripts locally with
whisper.cpp. - Manage session-level resources from the session detail page.
- Manage app-level resources, models, storage, runtime checks, and background tasks from Settings.
Architecture
- Desktop shell: Tauri 2.
- Frontend: React, TypeScript, and Vite.
- Backend: Rust commands exposed to the Tauri frontend.
- Persistence: JSON files for session data.
- Recent state: Tauri Store plugin.
- Transcription runtime:
whisper.cpp/whisper-cli. - Media runtime: bundled or locally resolved
ffmpeg.
Runtime model
Leclog keeps the main workflow local. Recording and imported media create local session files, and transcription jobs write generated artifacts back into app-managed storage.
Runtime checks in Settings report whether the app data directory is writable, whether ffmpeg and whisper-cli are available, whether a local Whisper model exists, and whether interrupted processing jobs or partial model downloads need attention.
Processing presets
The app exposes transcription processing presets:
Fast: smaller model preference and shorter chunks.Balanced: default local transcription preset.Accurate: larger model preference and wider overlap.Custom: manual chunk length, overlap, thread count, and refresh interval.
Release channel
Current releases target macOS Apple Silicon. The release workflow builds signed and notarized .app and .dmg artifacts, produces Tauri updater metadata, and publishes assets through GitHub Releases.
The latest published release is v0.3.5.