HeyZoku

Orchestrate an army of coding agents with your voice.

Mac

Productivity

Developer Tools

Visit WebsiteSee on Product HuntTwitter ⧉

the interrupt/redirect demo is impressive, but the part I keep coming back to is false positives - if it's always listening for agent-directed commands, how does it tell "claude, stop" meant for an agent apart from the exact same words coming up in a normal conversation with someone else in the room? that seems like the failure mode that actually matters once you're trusting it to interrupt live work mid-run, more than the discomfort of narrating commands out loud.

the interrupt demo is genuinely cool, but I'm skeptical this fits how most of us actually work. open-plan offices, coworking spaces, a partner on a call in the next room - narrating commands to a fleet of agents out loud isn't something I'd want to do most of the day, even with headphones on. feels like it's built for a solo-founder-in-a-home-office use case specifically, which is real but narrower than "voice is the future of how we interact with machines" suggests

The on-device Parakeet model is the detail that determines whether this can actually live in a dev workflow — cloud speech adds variable latency that breaks the flow of directing multiple agents. The design where an agent only surfaces when it needs you, rather than reporting continuously, is the right mental model for background parallelism. One thing I would want to understand before integrating into my setup: does HeyZoku keep a persistent index of what each agent is working on, or does the session state reset when I close and reopen the app? That determines whether I can pick up a multi-agent run the next morning or start fresh each time.

One thing that would make this killer for me: a visual dashboard showing which agent is working on what branch right now. When you have ten of them going at once, even naming them by voice can get confusing about what task each one actually owns. A simple grid view with live status and the last few spoken commands per agent would save you from having to query each one separately to figure out where things stand.

Spent a weekend with HeyZoku and the voice interrupt is honestly the killer feature - just saying "Claude, stop" mid-run and rerouting feels like sci-fi. Runs light on my M2 too.

Congrats on the launch! Letting a dev run a whole fleet of coding agents hands-free, completely offline, is a bold bet on where developer tools are heading. When multiple agents want your attention at once, how do you decide what actually gets surfaced to you versus queued?

Voice as the control layer makes sense for orchestration specifically — you're issuing intent, not typing code, and intent is the part that survives being spoken.

Where I'd expect it to get awkward is the return path. Agents running in parallel produce output in parallel, but speech is strictly serial. If three of them come back with questions at once, does it queue them, interrupt, or fall back to the screen for anything that needs a real answer? Feels like the hardest design problem in the whole thing isn't the listening, it's deciding what's worth speaking.

Congrats on the launch, @dangibuilds ! 🎉 Love the idea of making terminal agents truly voice-first, especially with offline processing using Parakeet.

One question: how do you handle situations where multiple CLI agents could interpret the same voice command differently? Is there an orchestration layer that routes commands to the right agent, or do users explicitly choose which agent they're talking to? Curious about the UX.