SKI

About SKI

SKI turns a coding agent into a conversation you can have out loud. You speak, the agent works, and it answers back in a natural neural voice — the loop is full duplex with echo cancellation, so you can interrupt it mid-sentence the way you would a person, rather than waiting through a completed response. Everything runs on the machine: speech-to-text is local, the voice is synthesised on-device, and the site's counters make the claim concrete at zero bytes uploaded and unlimited recording retention, because nothing is ever sent anywhere to retain. Setup is one installer — DMG on Mac, EXE on Windows — followed by a wizard that configures microphone, voice and a global hotkey, then one click per agent to connect. Six or more agents are supported including Claude Code, Cursor, Codex, Gemini CLI and OpenClaw, and the connection is a small skill the agent already understands rather than a proxy in front of the model. Beyond dictation it handles multi-project switching, agent liveness so you know whether the thing you are talking to is still running, an approve-before-send step so a misheard instruction does not reach the agent, and a meeting transcriber that reuses the same local speech stack. It is free for life with no credit card, which given the entirely on-device architecture means there is no per-minute cost to pass on.

Key Features

SKI Pros & Cons

Who Is SKI Best For?

Tags