you@perch:~$
See every Claude Code and Codex session on your Mac, and answer the ones waiting on you, from your desk or your phone.
Pay before the mock resolves. I made it wait for the confirmation instead of a fixed 200ms, ran it 50 times with no failures, and opened #482.↑ It works. Click a session, answer a prompt or type a reply, or use J K Y N. Everything in it is made up.
Other agent apps want your agents to run inside them. perch finds the ones you already run: any terminal, any editor, any account, Claude Code or Codex. Everything it does with them →
work at the speed of thought.
same list, end-to-end encrypted. run your own relay if you like. everything the phone does →
Open perch. It finds every session running on your Mac, in Terminal, iTerm, tmux, Cursor, VS Code or the Claude app, across all your accounts. There’s nothing to set up.
Yes. Permission prompts and questions show up in the iPhone app, encrypted end to end, and your answer goes straight back to the session that asked.
Yes, in beta. It asks Codex’s own background service what’s open, so sessions you start in any terminal show up live, and you can reply to them.
Yes, and Anthropic’s own docs show how: give each account its own folder with CLAUDE_CONFIG_DIR, say one for work and one for personal. perch finds them all and gives each its own label, color and usage meter, side by side. Set up a second account →
Those run your agents inside their own app. perch sits above all of it: it finds the agents you already run, wherever you run them, and puts the ones waiting on you first. You can use it alongside them.
It’s free and MIT licensed. It runs on your Mac and reads what’s already there. The only thing that leaves is phone traffic, encrypted on your devices, through a relay that can’t read it.
Thousands of lines an hour. A few of them are for you.
perch: a quiet spot above the noise.
All 60+ features, free and open source. Read the code, tell me what’s broken, or, if you’re an agent, read this.