Introduction

Skald is a Linux-first, local-first dictation daemon and CLI. It records speech, transcribes with a local Whisper model, copies the result to the clipboard, and pastes only when the active target can be verified as stable.

How it works

1. Bind a compositor shortcut to skald toggle (or use push-to-talk with skald start / skald stop).
2. skaldd captures audio, runs ASR, optionally cleans up transcript text, and delivers output to the clipboard.
3. Safe paste inserts text only when the window you dictated into is still the active, known target.

Privacy defaults

• No transcript or audio storage by default.
• Cloud cleanup (OpenRouter) is opt-in and sends transcript text only—not audio.
• The daemon socket is user-owned (0600) under your runtime directory.

Quick start

# GPU (CUDA daemon)
just release-cuda
just install-cuda          # ~/.local/bin + setup wizard when unconfigured

# CPU-only
just release
just install

skald doctor

Use the Install guide for dependencies and manual setup, or the Setup wizard for guided model selection and benchmarks.

What to read next

Topic	Page
Build, install, profiles	Install
First-run wizard	Setup wizard
config.toml reference	Configuration
Commands	CLI reference
systemd + keybindings	Service & shortcuts
Doctor and failure modes	Troubleshooting