Vux
Voice control for your agents.
Manage your agents on the go just by talking to them.
Open the link on your iPhone. It installs Apple's TestFlight app first if you don't have it, then Vux. The beta is open to the first 100 testers.
Features
- Streaming speech: replies play sentence by sentence as tokens arrive.
- Barge-in: talk over a reply to cut it off.
- Talk while it works: ask how it's going, say "stop", or change course without starting over.
- One step to start: the Action Button, Siri, Spotlight, or a Control Center or Lock Screen control.
- Voice approvals for commands the agent wants to run.
- Notifications when a task finishes, fails or needs approval while you're away (Hermes).
- Hands-free: runs in the background, with a Live Activity on the lock screen, Dynamic Island and CarPlay. On headphones, your music or video keeps playing.
- Reports back: background work tells you when it's done, in the conversation it came from. With a Vux skill like herdr-voice, the agents at your desk show on their workspace's page, with a card when one has a question or a result for you.
- Photos: attach up to 4 to your next turn.
- Result cards for PRs, issues, test runs, diffs and links. Long-press any message to copy it or its links.
- Sign in from your phone: open the link the agent sends, paste the code back in one tap.
- Every conversation, every computer: pick up any past conversation on the computer it ran on, give each computer its own model, hand off to Hermes Desktop.
- Vux skills: Skills that answer in a few spoken sentences and can add pages to swipe to, like herdr-voice, which gives each Herdr workspace a page with its Claude Code and Codex agents. Install them from settings, or just ask your agent.
- Your agent: Vux's own, built on pi-durable, or Hermes Agent if you have it.
- Private: talks only to your own computer and ModelRelay, or the voice provider you pick. No accounts, no tracking.
How it works
- You talk. Vux streams your voice to Grok's speech-to-text, so your words are ready for your agent when you stop.
- Your agent does the thinking on your computer, with your tools and skills.
- The reply streams back, and Gemini Flash speaks it sentence by sentence.
Why ModelRelay? Vux needs a speech-to-text model and a voice. ModelRelay serves both behind one key, from the team behind Vux. The setup command sets it up for you: you sign in from your phone, and new accounts come with free credit. Vux's own agent runs its model there too; Hermes keeps its own.
Your own voice? Settings › Voice also takes an OpenAI-compatible provider with your own key, such as OpenAI's. ModelRelay hears you while you talk, so replies come back quickest; OpenAI's voice starts a little sooner and handles code words better, but each turn waits for your recording to upload.
Get started
- Install Vux from the TestFlight beta.
- In Vux, tap Copy setup command.
- Paste it into an agent or Terminal on your computer.
- Tap Connect.
No config files or keys to copy.
Needs iOS 26, a Linux or macOS computer and Tailscale.
Setup · Vux skills · Privacy