Vocoflow mayankgujrathi

dictation rust speech-recognition speech-to-text

Use this command to install Vocoflow:

winget install --id=mayankgujrathi.vocoflow -e

Turn your voice into ready-to-paste text in seconds with a fast, privacy-friendly desktop dictation app built in Rust.

Vocoflow

Turn your voice into ready-to-paste text in seconds with a fast, privacy-friendly desktop dictation app built in Rust.

> ⚠️ Hobby project note: Vocoflow is currently a hobby project. To keep shipping features quickly, official code-signing/notarization is not fully set up yet.

Quick Start

Requirements

A working microphone (uses system default mic).

Install from GitHub Releases (recommended)

Open the GitHub repo’s Releases tab.
Download the latest build for your OS:
- Windows: vocoflow--windows-installer.exe
- macOS: vocoflow--macos.dmg
- Linux: vocoflow--linux.AppImage
Install/run using the platform steps below.

If your OS shows a security prompt while opening/installing, see Unsigned Install Guidance.

Package managers

winget

winget install -e --id mayankgujrathi.vocoflow

brew

Homebrew installs are available for macOS and Linux.

Primary (fully-qualified, recommended to avoid naming conflicts):

brew install mayankgujrathi/tap/vocoflow

Alternative:

brew tap mayankgujrathi/tap
brew install vocoflow

Notes:

macOS release builds are currently Apple Silicon only.
Windows users should install via winget.

Release Trust (End-user)

Releases are built and validated in CI across Windows/macOS/Linux.
Artifacts include integrity/safety checks (like checksum generation and automated security scans) to reduce risk.
Full technical details are documented in Build and Release.

Note: These checks improve trust, but no software distribution can promise zero risk.

Build from source

cargo build --release

Run from source

cargo run --release

Use

Press Ctrl + ` (Windows/Linux) or Command + ` (macOS) to start recording.
Speak normally.
Press the same hotkey again to stop.
Transcribed text is copied and typed into the active text field.

Unsigned Install Guidance (Windows/macOS/Linux)

Because this is a hobby project and binaries may be unsigned on some platforms, you may see OS security warnings.

Why this happens: full commercial-style code signing/notarization is still being rolled out.

What assurance you still get: release builds go through CI validation and automated security checks before publishing (see Build and Release for details).

Windows (Unknown Publisher / SmartScreen)

Launch installer.
If SmartScreen appears, click More info.
Click Run anyway.

macOS (Gatekeeper)

Try opening the app from Finder.
If blocked, right-click app → Open → confirm.
If still blocked, go to System Settings → Privacy & Security and click Open Anyway for Vocoflow.

Linux (AppImage)

Make the file executable:
```
chmod +x vocoflow--linux.AppImage
```
Run it:
```
./vocoflow--linux.AppImage
```

Tip: Always download from the official GitHub Releases page.

First Launch: What to Expect

The app may take a little longer the first time while it prepares runtime files.
A settings.json file and log files are created automatically.
You might not see a main window on startup — Vocoflow can run in the background. Check the system tray/menu bar icon to confirm it’s running and open Settings.
On first launch, you may briefly see a small blue progress pill. That indicates model files are being downloaded for transcription.
For model source and license details, see Attribution.
Depending on your OS, you may be asked to allow microphone/accessibility/security permissions.
After setup, start dictation with the hotkey and use normally.

Troubleshooting (Quick)

Windows webview issues: make sure WebView2 runtime is installed.
Linux webview issues: install WebKitGTK stack (webkit2gtk, javascriptcoregtk, libsoup3).
macOS WebKit link issues: build with RUSTFLAGS="-l framework=WebKit" when needed.

For full troubleshooting and platform-specific setup details, see the docs below.

LLM Integration (Optional, OpenAI-compatible)

Vocoflow includes optional LLM-based transcript post-processing for cleaner, context-aware output.

Why use it (advantages)

Produces cleaner punctuation/casing automatically.
Improves readability and polish for notes, docs, chats, and emails.
Adapts style to the currently focused app context (window/app metadata).
Can move from light cleanup to fully rewritten polished output depending on selected level.

API compatibility

Uses an OpenAI-compatible API endpoint (llm_base_url).
Works with providers/local gateways that expose OpenAI-style chat completion APIs.

Default behavior

By default, LLM reformatting is not enabled (reformatting level = none).
In none mode, model calls are generally bypassed.

Settings & Logging

To keep this README focused on install/use, detailed settings and logging behavior has moved to:

Settings and Logging

Documentation

License

MIT — see LICENSE.