SKI — Releases & downloads

Voice for your coding agent. Talk to Claude Code, Cursor, Codex, Gemini CLI and more — out loud — and hear them answer in a natural voice. Everything runs on your machine; nothing you say ever leaves it.

→ heyski.io

This repository hosts the official SKI downloads and the auto‑update feed for macOS, Windows and Linux. It contains release binaries only — not the app source.

Download

Get the newest build from the Releases page:

Install

macOS

1. If SKI is already installed, delete /Applications/SKI.app first — old copies can stay Gatekeeper‑blocked and a fresh download won't heal them.
2. Open the DMG and drag SKI to Applications.
3. First launch shows the standard "downloaded from the Internet" prompt → Open.
4. Verify: SKI → Preferences → About shows the version.

Windows

1. Run the installer. Windows may show a SmartScreen “Windows protected your PC” notice (the build isn’t code‑signed yet) → click More info → Run anyway.
2. Follow the prompts, then launch SKI from the Start menu.

Linux

1. Install the package — use apt, not dpkg -i, so dependencies resolve:

   sudo apt install ./SKI_<version>_amd64.deb
   

2. Launch SKI from your applications menu.
3. First run downloads a ~620 MB speech model. Transcription starts working once it finishes; the download resumes if interrupted and is checksum‑verified. The voice is bundled and needs no download.

Requires Ubuntu 24.04 or newer / Debian 13 or newer. Older releases ship a glibc SKI can't run on, so apt declines the install rather than leaving you with an app that won't start.

Verify your download

shasum -a 256 SKI_<version>_aarch64.dmg     # macOS
sha256sum SKI_<version>_amd64.deb           # Linux

The expected hash is printed in each release's notes.

Automatic updates

Once installed, SKI checks for new releases automatically. On macOS the update downloads in the background and applies when you relaunch; on Windows and Linux you click Restart now in Preferences → About and the signed package installs. Every update is cryptographically signed, and SKI verifies the signature before applying it — so even a compromised download host can never push a tampered build to you.

Make your agent talk — the SKI skill

The app gives you the widget; the SKI skill gives your coding agent ears and a voice. Install it once per agent — instructions are in that repo.

Security & privacy

macOS builds are Developer‑ID signed and notarized. Windows builds aren’t code‑signed yet, so Windows may warn on first install. Linux has no OS‑level code signing at all, so the published SHA‑256 and the update signature are the whole chain there. On every platform, each auto‑update is cryptographically signed and verified before it’s applied. Speech‑to‑text and the voice both run on‑device — your voice is never uploaded. Details at heyski.io.

Support

[email protected]