Image: SKI
Why it mattersA developer who runs several coding agents at once no longer needs the keyboard to steer each one. They answer the one that asked.
Running one coding agent fits in a terminal. Running six does not. SKI is a Mac app that puts six coding agents side by side in one window. The supported agents are Claude Code, Codex, Cursor, OpenCode, Antigravity and pi. The developer talks to it. It talks back. SKI reached number one Product of the Day on Product Hunt. It had 238 upvotes on 31 October.
What SKI does
SKI is a harness, not a model. The user brings the accounts they already pay for. SKI picks the engine for each task. Each chat gets its own git branch and its own worktree. The agents never touch each other's files. The developer hears the moment one of them needs an answer.
Voice is the main input. The site says SKI "works out the intent and asks instead of guessing". It does not pass raw transcript to the agent. The site also shows a risk case. A misheard "delete the test file" could run as "delete the rest of the file". SKI shows the heard words for review before each action. Hands-free use works while the developer paces or sketches. SKI auto-detects the language. The site shows Hindi, Spanish and Japanese examples.
What it does across projects
SKI runs more than one project at a time. The developer asks where things stand. SKI gives one spoken answer. It then hands the next job to the right agent. The site shows a case with two agents running on one project. A feature branch waits on hero copy. An API branch has green tests.
Git is built in. Every chat has its own branch and worktree. The developer can watch the diff as it is written. A spoken "open the PR" opens a pull request. The site shows a four-file commit on a hover-fix branch opened as PR number 142.
Pricing, platforms and tradeoffs
SKI runs on macOS. The site says Windows and Linux are "coming soon." The Free plan is zero dollars a month. It gives two sessions at once and 300 credits a month. The Pro plan is twenty dollars a month. It gives twelve sessions at once and 2,000 credits a month. Enterprise pricing is on request. The site does not say how a credit is counted. The developer still has to bring their own Claude, OpenAI, Cursor or OpenCode account.
The pitch rests on one claim. Review is the slow part of a day with many agents. A spoken summary and a yes-or-no answer beats scrolling six tabs. The site also admits the opposite risk. A spoken order can be heard wrong. An irreversible action needs confirmation. SKI shows the heard words before each action and asks before anything destructive.
A developer running one agent still has a terminal. SKI is for a different day. On that day five or six agents are at work. The main job is keeping track of which one needs the next answer.
Source
SKI website, by Anand Balakrishnan, Anoop Balakrishnan and John.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
