SKI provides a two-way voice loop for AI coding agents, supporting one-click integration with tools like Claude Code, Cursor, and Gemini CLI.
Setup is quick: a single installer (DMG for Mac, EXE for Windows) configures mic, voice, and hotkey in minutes without requiring a credit card.
All voice processing (speech-to-text and neural voice synthesis) runs locally on the device, ensuring no audio data is ever uploaded to the cloud.
The user speaks requests, which are transcribed on-device and sent to the coding agent; the agent’s replies are spoken aloud in a natural voice and displayed in a floating pill or notch UI.
Features include full-duplex echo cancellation allowing interruptions, silent mode for on-screen-only replies, and optional review mode with editable bubbles before the agent receives input.
The tool supports multiple simultaneous repositories, screen capture (via hotkey or agent-initiated), and integration with Google Calendar for agent participation in video meetings.
Video meetings can be recorded locally (free, on-device, unlimited) with speaker-tagged tracks, or agents can join via AgentCall (paid) for live participation with screensharing and chat.
All transcripts, voice sessions, and recordings are stored locally in a built-in Transcripts window, where users can view, edit, copy, export (Markdown or plain text), or delete data.
System requirements: Apple Silicon Mac (M1+, macOS 14.4+) or Windows 11/10; English is supported now, with more languages planned. Linux support is on the roadmap.
SKI is free forever for on-device features; the only paid feature is AgentCall for agent participation in calls, billed per minute with free hours included.