Helping you sound crispy.
Edit video by editing text — entirely on your machine.

Features · How It Works · Quick Start · Contributing

---

Why Toaster?

Recording yourself is easy. Editing out every "um", false start, and awkward pause? That's the hard part.

Toaster is a transcript-first desktop editor for spoken audio and video. Instead of scrubbing a timeline, you read your words, select the ones you don't want, and delete them — just like editing a document. Toaster handles the audio splicing, waveform sync, and caption export behind the scenes.

Everything runs locally. No cloud APIs, no uploads, no subscriptions.

Features

• Edit media by editing text — see your transcript, select words, delete/silence/restore in one click
• Local transcription — generate word-level transcripts with on-device models (Whisper ecosystem)
• Filler & disfluency detection — automatically highlight "um", "uh", "you know", and pauses
• Non-destructive editing — every action is reversible; your original file is never touched
• Synchronized playback — transcript, waveform, and video stay in lockstep as you edit
• Export cleaned media — render your final cut with captions (SRT/VTT) and script text
• Save & resume — project files preserve your edits for iterative sessions
• Privacy-first — no runtime network calls, no telemetry, fully offline

How It Works

1. Open a video or audio file
2. Transcribe with a local model — Toaster generates a word-level transcript
3. Read and edit — select words you want to remove and hit Delete
4. Preview — play back your edit in real time with synced waveform and video
5. Export — render the cleaned media plus captions and script

The entire workflow stays on your machine. Your media never leaves your computer.

Quick Start

Install from release

Download the latest installer from the Releases page.

Platform	Architectures	Format
Windows	x64, ARM64	.msi
Linux (Debian)	x64, ARM64	.deb
Linux (RPM)	x64, ARM64	.rpm
Linux (any)	x64, ARM64	.AppImage

Note: Windows installers in v0.1.0 are unsigned — SmartScreen will show "Windows protected your PC" the first time. Click More info → Run anyway to install. Code signing is planned for a follow-up release.

macOS builds are not currently published. Build from source if you need a macOS app — see docs/build.md.

Build from source

See docs/build.md for full platform setup. The short version:

bun install --frozen-lockfile
cargo tauri dev

On Windows, run .\scripts\setup-env.ps1 first to configure the MSVC + LLVM build environment.

Troubleshooting

"localhost refused to connect" on launch (Windows)

If the installed app shows a localhost refused to connect page instead of the editor, you most likely have a leftover debug build of toaster.exe still running from a previous cargo tauri dev / cargo tauri build --debug session. Tauri's single-instance plugin forwards every Start-Menu launch to that stale window, which expects a Vite dev server at http://localhost:1420 and shows the WebView2 connection error when it doesn't answer.

Get-Process toaster | Stop-Process -Force
Start-Process "C:\Program Files\Toaster\toaster.exe"

The released MSI from the Releases page bundles all frontend assets directly into toaster.exe and never contacts localhost:1420.

Tech Stack

Layer	Technology
Desktop shell	Tauri 2.x
Backend	Rust
Frontend	React · TypeScript · Tailwind CSS
State	Zustand
Transcription	Local model inference (Whisper ecosystem)
Export	FFmpeg 7

Contributing

We welcome contributions! Please read CONTRIBUTING.md before opening a PR.

# Run the checks contributors are expected to pass
cd src-tauri && cargo test && cargo clippy
npm run lint

For translation contributions, see CONTRIBUTING_TRANSLATIONS.md.

Acknowledgments

Toaster is forked from Handy by CJ Pais. Handy proved that a free, open-source, fully-offline speech tool could be simple, private, and community-driven. Toaster builds on that foundation with a transcript-first editing workflow.

We're grateful to the projects that make Toaster possible:

• Tauri — the Rust-native app framework that keeps the bundle small and the runtime fast
• Whisper by OpenAI — the speech recognition model at the heart of local transcription
• whisper.cpp & ggml — cross-platform inference and hardware acceleration
• FFmpeg — the Swiss Army knife of media processing

License

MIT — see LICENSE for details.