Launching today
SKI
Free voice coding for Claude Code, Codex and more
734 followers
Free voice coding for Claude Code, Codex and more
734 followers
We just built SKI — voice coding for Claude Code, Codex and more. It's not dictation: your agent answers you out loud, like a real teammate, so you build at the speed you think. You can even bring it into a meeting to build live, or send it in your place to speak for you. It's an ambient thing that just sits on your desktop — hit a key, talk, it works. All on your machine, free. Available on Mac & Windows.
Yes, I want this! Giving it a try.
Heads up: Clicking the download buttons on your site did not work for me (but copying the link manually did)
Update - initial feedback:
- Very quick to setup and understandable! Love how it just instantly works on my existing setup without any configuration.
- The talk-indicator doesn't feel responsive. It needs three important visual states: recording started (key pressed), listening, recording ended (key released)
- Since push-to-talk doesn't present anything visually on screen, I find it hard to stay focused on the audible conversation. The effect is something like taking a regular phone call while working, as opposed to having a video call with screen sharing. While it's very neat to have it out of the way, I think SKI might work better if it had a fuller GUI showing what's being talked about. Maybe a 400x300 floating window where SKI can pull up the current artifact (current plan, PR description, commit log, quickly sketched diagram, screenshot, etc)
- What SKI calls "Projects" are actually worktrees/agents within a project. Maybe not crucial now, but might be worth a retake (either call it workspaces, or introduce a hierarchy).
Question:
Can I switch project by voice? (I'm afraid to ask, if my agent ends up trying a creative solution...)
SKI
@erik_v Thanks a lot for the detailed feedback. Fixed the download issue. Thank you for letting us know.
Will work on the indicator feedback. At present we have only focused on listening, thinking and speaking modes. Will add more indicators for the states of recording in the upcoming update.
We are working on a little more complex version of ski, like you have mentioned, where in, SKI gets more visual indicators and controls compared to the subtle presence it has now.
Unfortunately, we haven't implemented an intelligence in existing SKI, and it will not be able to switch project by voice. It is hotkey controlled. But if you ask it to get something from another project, and if the agent connected to SKI has access to the folder, it will bring that in for you. But the context might get mixed up. Will come up with a solution in the next version :)
Block Convey Drive
SKI
@arun_prasad06 At present, it is a skill available for agents like Claude Code, codex or hermes. It can be used with any agents that are capable to use a skill to execute a task. It is different from Whisprflow, which is a transcription application, converting speech to text. SKI is more like a skill which you agent has. SKI automatically hands it over to agent, and the agent then does a task and tells SKI what to say out loud.
claude code, and the thing that would keep it open is voice for describing intent but never for reading the diff back. those are different jobs and only one of them wants a mic
SKI
@alex_watson2110 Yes, voice is mostly for conversations and status, and less for read diff aloud.
Hi,
What languages are supported?
SKI
@oleksandr_utkin Great question! Right now @SKI is tuned for English and it works really well. Support for more major languages is next on our roadmap — which one would you want first? 🙌
SKI
@rahul_saji007 Thank you for checking our product.
Yes, the downloader will fetch the transcription model files (which you can choose), when on-boarding the project. Once it is downloaded, there is no requirement for internet for the application to work. Only the initial downloads need internet connectivity.
SKI
@thomas_digaetano The transcription usually works well with most accents. But whenever they go wrong, LLMs are intelligent enough to correct them, just like they correct the spelling mistakes we have when we type.
Small feedback upfront, when I simple click on "Download for Mac" on the website, the download on my device doesn`t start. Workaround found, if I do right click and open in a new tab, then it start the download.
SKI
@nik_isonpase Thanks a lot for letting us know. Will fix it soon.