SKI is a voice-first coding harness that lets developers manage multiple AI coding agents simultaneously using spoken instructions and real-time audio feedback.
What the product does and how it is positioned
SKI functions as an intelligent orchestration layer designed to address the bottleneck of managing multiple asynchronous AI coding agents in parallel. Rather than requiring developers to constantly monitor screens and type commands, the application provides a hands-free voice loop that speaks updates and requests clarification when an agent needs user input.
The platform operates neutrally across various third-party models and agents, including Claude Code, Codex, and OpenCode. It integrates directly with Git by assigning an isolated worktree and branch to every chat session, enabling developers to review diffs and open pull requests entirely through voice commands while keeping all code stored locally on macOS.
Source-supported ways to use the product
Managing several coding agents working on different tasks simultaneously, listening for audio prompts when an agent requires human intervention.
Creating branches, reviewing live code diffs, and opening pull requests using spoken commands during development.
SKI allows developers to run multiple AI coding agents side by side under a single interface, assigning distinct tasks to different engines based on complexity or cost. A built-in orchestration layer maintains project context across branches and chats. Additionally, users can set up agent chains where one agent plans a task and automatically hands the output over to another agent for building, all tracked through voice summaries.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
Typing text chats to agents is free for life with unlimited usage on the free plan, while real-time voice features operate on session limits and credit allowances.
The full harness is currently available for macOS, with Windows and Linux versions listed on a waiting list.
Your projects, code, and chats stay locally on your Mac, and your agents run on your own configured accounts.
Yes, the neutral harness supports various agents and models such as Claude Code, Codex, Cursor, OpenCode, and Antigravity side by side.
The harness reconstructs intent from project context rather than relying solely on raw transcription, and gates destructive or irreversible steps until confirmed.
