Simple steps to start Voice Coding with SKI

Thanks a lot for supporting and for being in the SKI community. This is a guide to get started, just in case you are stuck somewhere.

This guide covers how to get started with voice coding. This guide can be used to connect projects which you already are working on, to SKI.

1. Download SKI (if you haven't already)
   Currently available on Mac (M series) and Windows. Will be launching soon on Linux.

2. Start SKI
   You will be shown an on-boarding process. For transcription, use "Ultra" for best English transcription accuracy and speed. Use "High" for multi-lingual support.
   Ensure that you install the skills for your agent in the process. The skill installation is mandatory for the agent to work.

3. Once the skill is installed, start the SKI application. It will show up in the notch or as a floating widget (depending on your preference).

4. Now, the next step is to connect your project to SKI. Go to your coding session (where you type the instructions to give inputs to the LLM coding agent - such as the prompt area inside Claude code or codex) and just type "SKI".

5. The agent will read the skill files (which we have already installed while on-boarding) and learn how to connect to SKI. It will run a python script and connect to SKI. Once this happens, and the connection is established, the project name and a green dot will come up on the SKI widget or notch. The green colour represents that the project is connected and live.

6. You can now talk to SKI. Whatever you talk will be read by the agent, the same way as you typed. It knows that the words are being transcribed from speech.

7. To mute, you can either press "Fn" key in Mac OS, or space key if the widget is selected or use hotkeys. You can also use the mute button, or click waveform to mute and unmute.

   
   You can connect multiple projects, and projects from different agents in the same way, at the same time. You can switch between projects using hotkeys or using mouse by clicking and selecting the project.

   
   Note that the agent will not be reading out all the output it generates. The agent knows that SKI is a way to talk to you. So, it will decide and tell SKI on what to update to user using audio. You can just ask the agent (by voice on SKI) on what updates it should speak out and the agent will adapt accordingly.
   

A simple video tutorial is available at :

Let us know if you face any issues or are stuck somewhere and we will be happy to help you!