Run all your coding agents at once and talk to them. They ask out loud, you answer by voice, nothing merges without your review. Works with Claude Code, Codex, Cursor and more, on the plans you already have.
Talk to your coding agents and hear them talk back. SKI runs Claude Code, Codex, Cursor, OpenCode, Antigravity and pi side by side on your Mac, and tells you the moment one needs you.
Thanks a lot for supporting us and for your feedback with SKI.
We have taken in your feedback and requests. We have transformed SKI into a voice-first coding harness, with enhancements to support multi modal inputs in realtime
Supports 85+ languages: You could speak to your coding agents in languages that you are comfortable in, and the agent will talk to you in the same language as well.
Prompt processing: It runs your audio and brings in your thoughts as a proper prompt, refined for most agents (with optional approval)
All agents in one UI: Brings in all agents into a single UI, with most commonly used tools for coding. Switch sessions and agents by voice.
Chained agents: With a single voice command, you can give the agent a chained task. For example, ask Opus 5.5 to plan, and then to hand it over to GPT Luna to code and then to Fable for review, in just a simple voice command
Intelligent router: With "One chat for all" turned on, you can just talk to a single agent and it will intelligently route to sessions (with optional approval)
Expressive voice: A more human-like, low latency voice for each sessions with 28+ voices
First of all, a huge thank you to this amazing community. Your support, curiosity, and willingness to try new things mean the world to us, and we're genuinely grateful to have you along on this journey.
The voice-first approach looks really interesting. Being able to talk to multiple coding agents and hear their responses without constantly switching between tools could make development much smoother. How well does it handle multiple agents working on different tasks at the same time?
What needs improvement
It would be great to have more control over multiple agents working at the same time, along with clearer task tracking and response management. Better voice recognition in noisy environments and more customization for voice responses would also make the experience even better.
vs Alternatives
I also considered using coding agents individually through tools like Cursor, Claude Code, and Codex. But SKI seems interesting because it brings multiple agents into one voice-first workflow, so I don’t have to keep switching between different tools.
In July we launched SKI as a voice layer for coding agents. It went to #1 that day, and the people who found us through that launch gave us the feedback that shaped what we are showing today.
One pattern kept coming up: the people who got the most out of SKI were running three or four agents at once and using voice to keep track of them. So we rebuilt SKI around that.
SKI is now a voice-first harness for coding agents:
• Run Claude Code, Codex, OpenCode, Antigravity, Cursor and pi side by side. Each chat gets its own git worktree, so agents never step on each other.
• Talk to any of them. Click the notch to move your voice to another chat. SKI reads the agent's question out loud and you answer by voice.
• Say "stop" and it stops. Undo a turn and the agent forgets it.
• 85+ languages. Speak the way you think.
• Every change waits on screen for your review before it goes anywhere.
• Your agents run on the subscriptions you already have. SKI only charges for its voice, and there is a free plan every month.
Now on Mac. Linux and windows coming soon. Sign in at justski.it.
What we would love from you: try it on a real repo and tell us what you said and what came back, good or bad. We are here all day.
— Anand, Anoop and John
Report
@anand_balakrishnan5 Congrats on the launch! Mac-only for now, or is Windows parity close?
@shivam_kushwaha16 If it is in the same coding agent, for instance, Opus to Sonnet, it hands over the session. If it is from a different coding agent, for instance, Opus to OpenAI models, it hands over the session and files seamlessly
@anand_balakrishnan5 can I converse in indic languages as well? And how much structured my voice input should be? What happens if my voice command is not optimal and ambiguous? Can it handle that?
This is awesome. Congrats on the launch. Just want to know what is the difference between using the voice feature on Claude Code, Codex or Cursor or OpenCode directly versus this?
This works across harness. Say, once voice can talk to Codex and ask it to hand over to Claude code and maybe to opencode etc.
It works with different sessions at the same time - There is a one voice for all agents option - You could talk to all agents in one voice. Or you can switch between sessions - who identify by their voice
It is bi-directional voice, not dictation. The agent can choose to talk back.
There is option for auto prompt processing - from your audio, SKI can create prompts in English - from your speech, which then goes into the agents
Multi lingual support - 85+ languages
This stays in notch on Mac (with optional hub) and can also be hidden. So, you can work on things while SKI runs
Detachable task bar - You will have a floating task bar which shows agents running, tasks in progress and which will just let you talk to it easily.
@tushaarmehtaa Thanks Tushar! One more difference: most of us pay for more than one agent but end up using just one. With SKI they all sit in one place, so every plan you pay for actually gets used.
@arjun_sureshh Not yet, and it is the next thing we are building. A SKI mobile app is in development, so you can check on your agents and talk to them from your phone, away from your desk. If that is the thing you want most, say so here; it helps us order the list.
Report
@anand_balakrishnan5 A mobile app would be really useful, Looking forward to it!
@arjun_sureshh Thanks Arjun! We'll share it here as soon as there's something to try.
Report
clicking the notch to move your voice to another chat is a neat detail :) if two agents ask something at the same moment, which one gets read out first?
@neerajnathany Adding to John's answer: in one-voice mode, SKI also says the project and branch name before the question, so you always know who's asking.
Report
Does it work with Claude Code on a Max plan, or only API keys?
@rahul_saji007 Yes. Your current Max plan works as is, no API key needed.
You can bring all your agent subscriptions into SKI. And if you have more than one Claude plan, connect them all: when one runs out of credits, switch to the next and keep going.
@gigi_kuriakose Yes, definitely. You can type to your agents just like in any other coding harness; every chat has a text box. Voice is the default, not the only way.
@viktor_radelytskyi We have built it from ground up. We had a completely local version before this, launched 3 months ago. We actively collected feedback from users, and came up with this version of SKI. The landing page is designed by @anoopbalak
@viktor_radelytskyi Thank you Viktor! Fun fact: the launch film was made in code too. Every frame is drawn by our own pages, and the dinosaur and MacBooks were rendered in Blender.
Report
@anand_balakrishnan5 That's actually interesting. What agent did you use? Did it still require any manual editing or was it rather straightforward?
Hi Product Hunt 👋
Anand here, one of the makers of SKI.
In July we launched SKI as a voice layer for coding agents. It went to #1 that day, and the people who found us through that launch gave us the feedback that shaped what we are showing today.
One pattern kept coming up: the people who got the most out of SKI were running three or four agents at once and using voice to keep track of them. So we rebuilt SKI around that.
SKI is now a voice-first harness for coding agents:
• Run Claude Code, Codex, OpenCode, Antigravity, Cursor and pi side by side. Each chat gets its own git worktree, so agents never step on each other.
• Talk to any of them. Click the notch to move your voice to another chat. SKI reads the agent's question out loud and you answer by voice.
• Say "stop" and it stops. Undo a turn and the agent forgets it.
• 85+ languages. Speak the way you think.
• Every change waits on screen for your review before it goes anywhere.
• Your agents run on the subscriptions you already have. SKI only charges for its voice, and there is a free plan every month.
Now on Mac. Linux and windows coming soon.
Sign in at justski.it.
What we would love from you: try it on a real repo and tell us what you said and what came back, good or bad. We are here all day.
— Anand, Anoop and John
@anand_balakrishnan5 Congrats on the launch! Mac-only for now, or is Windows parity close?
@fedulovelena Thank you, Helen! We are working on the Windows and Linux versions now, and we will let you know as soon as they launch.
@anand_balakrishnan5 Can it switch between agents smoothly while keeping the full context
@shivam_kushwaha16 If it is in the same coding agent, for instance, Opus to Sonnet, it hands over the session. If it is from a different coding agent, for instance, Opus to OpenAI models, it hands over the session and files seamlessly
@anand_balakrishnan5 can I converse in indic languages as well? And how much structured my voice input should be? What happens if my voice command is not optimal and ambiguous? Can it handle that?
This is awesome. Congrats on the launch. Just want to know what is the difference between using the voice feature on Claude Code, Codex or Cursor or OpenCode directly versus this?
@tushaarmehtaa
This works across harness. Say, once voice can talk to Codex and ask it to hand over to Claude code and maybe to opencode etc.
It works with different sessions at the same time - There is a one voice for all agents option - You could talk to all agents in one voice. Or you can switch between sessions - who identify by their voice
It is bi-directional voice, not dictation. The agent can choose to talk back.
There is option for auto prompt processing - from your audio, SKI can create prompts in English - from your speech, which then goes into the agents
Multi lingual support - 85+ languages
This stays in notch on Mac (with optional hub) and can also be hidden. So, you can work on things while SKI runs
Detachable task bar - You will have a floating task bar which shows agents running, tasks in progress and which will just let you talk to it easily.
@tushaarmehtaa Thanks Tushar! One more difference: most of us pay for more than one agent but end up using just one. With SKI they all sit in one place, so every plan you pay for actually gets used.
Can I use it away from my desk, from my phone?
@arjun_sureshh Not yet, and it is the next thing we are building. A SKI mobile app is in development, so you can check on your agents and talk to them from your phone, away from your desk. If that is the thing you want most, say so here; it helps us order the list.
@anand_balakrishnan5 A mobile app would be really useful, Looking forward to it!
@arjun_sureshh Thanks Arjun! We'll share it here as soon as there's something to try.
clicking the notch to move your voice to another chat is a neat detail :) if two agents ask something at the same moment, which one gets read out first?
@neerajnathany The one which you are currently working on gets the first priority as you are engaging with it actively.
@neerajnathany Adding to John's answer: in one-voice mode, SKI also says the project and branch name before the question, so you always know who's asking.
@rahul_saji007 Yes. Your current Max plan works as is, no API key needed.
You can bring all your agent subscriptions into SKI. And if you have more than one Claude plan, connect them all: when one runs out of credits, switch to the next and keep going.
@anand_balakrishnan5 @rahul_saji007 Thank you for checking out SKI.
Please try it out and let us know if you face any issues with the product. Happy to help!
Can I still type when I don't want to talk?
@gigi_kuriakose Yes, definitely. You can type to your agents just like in any other coding harness; every chat has a text box. Voice is the default, not the only way.
@anand_balakrishnan5 Thank you
@gigi_kuriakose Anytime, gigi! Tell us how it goes once you've tried it.
This demo as well as the landing page looks pretty smooth, ngl. How exactly did you build it?
@viktor_radelytskyi We have built it from ground up. We had a completely local version before this, launched 3 months ago. We actively collected feedback from users, and came up with this version of SKI. The landing page is designed by @anoopbalak
@viktor_radelytskyi Thank you Viktor! Fun fact: the launch film was made in code too. Every frame is drawn by our own pages, and the dinosaur and MacBooks were rendered in Blender.
@anand_balakrishnan5 That's actually interesting. What agent did you use? Did it still require any manual editing or was it rather straightforward?