SKI documentation
It works with Claude Code, Codex, Cursor, Gemini CLI, Windsurf and OpenClaw, on macOS and Windows, and it is free for life.
Get the app running on Mac or Windows in two minutes
Type ski in your agent and watch the dot turn green
Speak a request and hear the answer come back
The widget, the skill, and the files in between
Why closing the loop changes how you code
What stays on your machine and what never leaves it
Bind several repos at once, each with its own voice
Mute, silent mode, screenshots, push-to-talk
Review every transcript before it reaches the agent
Unlimited on-device recording with a live transcript
Your agent joins the call as a visible participant
Send your agent into recurring meetings on time
What SKI is not#
SKI is not a dictation tool. Dictation stops at text-in: you speak, words appear, and you are back to reading a screen. SKI closes the loop. Your agent's reply is synthesized locally and spoken aloud, and full-duplex echo cancellation means you can cut in mid-sentence and still be heard.
SKI also does not type into your editor or terminal. It talks to your agent through a small skill and plain files inside your own project, which is why it works with any terminal, editor, or IDE you already use.
Where to go next#
New here? Start with Install SKI. Curious how the pieces fit before you download anything? Read How SKI works.