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.
Good to see SKI at #1 today. Running Claude Code, Codex and Cursor by voice is useful, and review before merge is important. Can one voice session pass context from one agent to another?
Yes. In normal voice mode, a session, for instance, running on Opus 5.5 can run the plan or maybe part of a code, and hand over the session to maybe Sonnet or GLM 5.3 Flash or others, as you request. If the models are from the same provider, they can carry over the conversation as well. If not, they carry over the intelligence.
Please try it out and let us know your feedback on the product. Happy to help!
Congrats on the launch. When an agent asks for a yes or no by voice, is the diff still shown on screen before anything is applied to the worktree? And in one-voice mode, how does it tell you which agent is asking when two are waiting at once?
The agent will explain the changes if required and show it up in a html file in Chrome visually, or you could also see it in the SKI hub as diff, just like the normal coding agents. The voice is intelligent enough to adapt according to your preferences.
In one-voice for all agents mode, when there are two agents, it will mention the branch name and project name to identify. In normal mode, each agent has its own voice and the selected session will be shown in the notch.
Please try out SKI and let us know your feedback too.
@akarsh_hegde Thanks Akarsh. To be clear on the main point: nothing gets applied without you seeing it first. Voice is just how you talk to the agents, and the review step stays the same.
@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?
The per-chat git worktree detail is what caught my eye — running Claude Code, Codex, and Cursor side by side only works if they never step on each other. With voice as the input, the review-before-anything-goes-anywhere step feels essential, not optional. Does saying “stop” pause just the chat I’m talking to, or all of them at once?
The review before send feels like a second check, just in case the user wants something to be corrected. Saying "stop" just stops the chat that you are in, and not all of them. SKI is relatively seamless and hands-free, designed to work with voice or multi-modal inputs whenever possible.
Please try it out, and we believe that you will feel it very comfortable and fast to build.
@jcbuilds Thanks JC! One small addition: voice is the default, but you can type in any chat whenever you'd rather not talk. Stop only affects the chat you're in, so the others keep going.
@rahul_saji007 In normal mode, each agent will have their own distinct voice. The session name will also be spoken if needed. In one-agent mode, the agent will tell you the name and details of the session which needs your attention.
Please try out SKI and let us know your feedback.
Report
Why launch again instead of just updating the old listing?
@rahul_saji007 SKI has undergone significant changes in its core fundamentals. While the first version of SKI was completely local, with slightly robotic voice with high latency, the new version has cloud components with more realistic voice at low latency. The new version is also intelligent and has improved in the user experience. We also have additional grounded voice-specific harness tools inside SKI, which makes the voice experience better.
Running four different coding agents at once and hearing about it the second one needs me is the right shape for this, since the real cost is babysitting tabs and never knowing when to switch back.
@karimbenkeroum Yes. And you can just work across agents (not just sessions). For instance, in Claude Code, all agents are models from Anthropic. In SKI, you could let different agents run together, side by side, and even let agents hand over their work to the next agent by voice.
congrats on the launch anand. the worktree per chat is the part that got me, i always lose track which agent touched what. one thing im curious, when the agent reads a long change back by voice how do you keep it short enough that i still really review it and not just say yes
The voice agent usually summarizes the changes in brief audio duration. If you want it to speak more or explain, it will. And it will bring up visual webpages and explain so that you can stay focused if needed. You can customize the behaviour as you like.
@anand_balakrishnan5 that makes sense. i am still curious about the voice part, is the read back short by default or do you have to ask for the short one
git worktrees solve the file-collision problem between agents, but what about state outside git - a shared dev server, a database migration one agent runs mid-task while another assumes the old schema. with 3-4 agents going at once that's the conflict I'd actually worry about, not two agents editing the same file.
Replies
Good to see SKI at #1 today. Running Claude Code, Codex and Cursor by voice is useful, and review before merge is important. Can one voice session pass context from one agent to another?
@modiyilreji Thank you for checking out SKI.
Yes. In normal voice mode, a session, for instance, running on Opus 5.5 can run the plan or maybe part of a code, and hand over the session to maybe Sonnet or GLM 5.3 Flash or others, as you request. If the models are from the same provider, they can carry over the conversation as well. If not, they carry over the intelligence.
Please try it out and let us know your feedback on the product. Happy to help!
@johnkg003 Thanks, John. The cross-model handoff is useful to know.
Congrats on the launch. When an agent asks for a yes or no by voice, is the diff still shown on screen before anything is applied to the worktree? And in one-voice mode, how does it tell you which agent is asking when two are waiting at once?
@akarsh_hegde Thank you for checking out SKI
The agent will explain the changes if required and show it up in a html file in Chrome visually, or you could also see it in the SKI hub as diff, just like the normal coding agents. The voice is intelligent enough to adapt according to your preferences.
In one-voice for all agents mode, when there are two agents, it will mention the branch name and project name to identify. In normal mode, each agent has its own voice and the selected session will be shown in the notch.
Please try out SKI and let us know your feedback too.
@akarsh_hegde Thanks Akarsh. To be clear on the main point: nothing gets applied without you seeing it first. Voice is just how you talk to the agents, and the review step stays the same.
Quick demo: SKI isn't only for code. I described a lounge chair out loud, and Claude Code modeled it in SketchUp step by step.
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?
The per-chat git worktree detail is what caught my eye — running Claude Code, Codex, and Cursor side by side only works if they never step on each other. With voice as the input, the review-before-anything-goes-anywhere step feels essential, not optional. Does saying “stop” pause just the chat I’m talking to, or all of them at once?
@jcbuilds Thanks a lot for checking out SKI
The review before send feels like a second check, just in case the user wants something to be corrected. Saying "stop" just stops the chat that you are in, and not all of them. SKI is relatively seamless and hands-free, designed to work with voice or multi-modal inputs whenever possible.
Please try it out, and we believe that you will feel it very comfortable and fast to build.
@jcbuilds Thanks JC! One small addition: voice is the default, but you can type in any chat whenever you'd rather not talk. Stop only affects the chat you're in, so the others keep going.
@rahul_saji007 In normal mode, each agent will have their own distinct voice. The session name will also be spoken if needed. In one-agent mode, the agent will tell you the name and details of the session which needs your attention.
Please try out SKI and let us know your feedback.
@rahul_saji007 SKI has undergone significant changes in its core fundamentals. While the first version of SKI was completely local, with slightly robotic voice with high latency, the new version has cloud components with more realistic voice at low latency. The new version is also intelligent and has improved in the user experience. We also have additional grounded voice-specific harness tools inside SKI, which makes the voice experience better.
Running four different coding agents at once and hearing about it the second one needs me is the right shape for this, since the real cost is babysitting tabs and never knowing when to switch back.
@karimbenkeroum Yes. And you can just work across agents (not just sessions). For instance, in Claude Code, all agents are models from Anthropic. In SKI, you could let different agents run together, side by side, and even let agents hand over their work to the next agent by voice.
Please try it out and let us know your feedback.
congrats on the launch anand. the worktree per chat is the part that got me, i always lose track which agent touched what. one thing im curious, when the agent reads a long change back by voice how do you keep it short enough that i still really review it and not just say yes
@agshin_rajabov Thank you for checking out SKI.
The voice agent usually summarizes the changes in brief audio duration. If you want it to speak more or explain, it will. And it will bring up visual webpages and explain so that you can stay focused if needed. You can customize the behaviour as you like.
@agshin_rajabov Thanks Agshin! Glad the worktree per chat clicked. Knowing which agent touched what was one of the first things we wanted to fix.
@anand_balakrishnan5 that makes sense. i am still curious about the voice part, is the read back short by default or do you have to ask for the short one
git worktrees solve the file-collision problem between agents, but what about state outside git - a shared dev server, a database migration one agent runs mid-task while another assumes the old schema. with 3-4 agents going at once that's the conflict I'd actually worry about, not two agents editing the same file.