Launched this week
Agent Clan
Command your Claude Code and Codex agents with gestures
86 followers
Command your Claude Code and Codex agents with gestures
86 followers
Agent Clan puts your Claude Code and Codex agents on your screen as a 3D team you run with your hands. Make a fist and they appear. Pinch and pull toward one to pick it, then just say what you need. When an agent asks for permission, pull right to allow or left to deny. Circle anything on your screen to hand it over, or ask a leader and its crew splits the work. Hand tracking and voice run on your computer, and everything also works with a mouse. Free on Mac and Windows.
Free Options
Launch Team
Render Sandboxes: a computer for every agentRun untrusted code safely. Idle sandboxes cost nothing
Promoted
Hi Product Hunt, I'm the developer of Agent Clan.
I kept switching windows to check on my Claude Code sessions while I was watching or reading something else. So I turned each agent into a small 3D character I can run with my hands.
How the gestures work:
Open your hand, then make a fist: your agents appear
Pinch and pull toward one to pick it, then pull up and just talk
When an agent needs your permission, its head pops up over your other apps. Pull right to allow, left to deny
Circle anything on your screen and hand it to an agent
Hand tracking (MediaPipe) and speech-to-text (Whisper) run on your computer, and camera frames never leave it. Agent Clan runs your own Claude Code or Codex in your project folder, and everything also works with a mouse and keyboard.
It's free on Mac and Windows. I'd love to know which gesture feels awkward, and what you'd want to do with your hands next.
Agent Clan is an independent product, not affiliated with Anthropic or OpenAI.
@_tsetse_ this is so cool!
Does the head that pops up show which command it wants to run?
@leoheo Yes, it says what it’s about to do!
@_tsetse_ Thanks! as text, or does it actually speak?
the permission pop-up over your other apps where you pull right to allow is a neat touch. does the hand tracking still work okay in dim light?
Hey !
This look like so cool !
Keeping an eye on several Claude Code sessions at once is a real pain point, so the agent's head popping over other apps when it needs permission is a smart touch.
Question: does it only work with local CLI sessions, or could it follow sessions running in the cloud too? Most of mine run on the web.
And for your next gesture: a swipe to send the same instruction to the whole crew would be handy.
The permission gesture is the part that stuck with me — pull right to allow, left to deny, with the agent's head popping up over whatever I'm doing. That's the exact moment I usually lose track of a run, so keeping it in view instead of buried in a terminal tab feels right. My one worry is the autopilot problem Xavier raised: at five sessions I'd be pulling right without reading. If the popup showed the command and the files it touches together, I'd trust it a lot more.
Pull right to allow and left to deny is a fun idea, and honestly the permission step is where I lose the most time babysitting agent sessions. How do you guard against an accidental approve when I'm just talking with my hands or reaching for coffee? Is there an extra confirm for riskier commands like a force push?
@anirudh_madhavan i think it won’t get through all of that by accident. To give a task, you need make a gesture to bring up the menu, pinch and pull to pick the agent, pull up for the mic, speak, then pull up again to send. Allowing works the same way, open the menu, pick the agent, then pull right and hold still for half a second while a ring fills. and Allow only shows up when that agent is actually waiting.
I also tried it at a family big table. The camera could see several people, all talking with their hands, and it still held up. other people’s hands don’t take over.
I think the best setup is one where you decide for yourself how to call things up, so user can record moves you’d never make by accident. Some people will probably want Naruto-style hand seals, and I’m all for it.
Two sessions, I read every command. Five sessions, I'm pulling right on autopilot. When a leader splits the work across a team, can I see which files each agent touched before I approve?