3 ms·
Great to see. It’ll be great to democratize access to remotely using cli coding agents. I’ve been iterating the past few months on a solution to use Claude Cod
by cadamsdotcom 8mo ago
Great to see. It’ll be great to democratize access to remotely using cli coding agents.
I’ve been iterating the past few months on a solution to use Claude Code on my phone while it runs on my laptop and it’s a lot of moving parts: Tailscale, git worktrees, tmux, an always-on “caffeinate” process, and a ton of hooks & tweaks to fix bugs along the way. It’s become very comfortable but in the process, impossible for anyone but me to understand.
But it’s awesome because I own the machine that runs tests and am not paying monthly for anything but Claude Max - and it keeps going if I lock my phone or go into a cell reception dead zone.
Productising such a thing would be a very interesting challenge indeed.
- groask 8mo agoTotally relate to the "tons of hooks & tweaks" part. I've been using Claude Code daily to build a macOS app solo, and ended up in the same rabbit hole — hooks, FSEvents, process monitoring, all duct-taped together. The remote/phone side I haven't solved (your Tailscale+tmux setup sounds solid for that), but the "at my desk" problem — losing track of which terminal is doing what — I ended up productizing into a small menubar app called GroAsk (groask.com). It does two things: ⌥Space to quick-launch any AI, and a dashboard showing real-time status of all Claude Code sessions so you can jump to the right terminal instantly. Very different scope from Omnara — purely local, purely desktop, no remote/phone stuff. But for the "I have 3 sessions running and forgot which one needs me" problem it's been a huge quality of life improvement for my own workflow.
- cadamsdotcom 8mo agoNice. For the “which session needs me” challenge I arrange windows spatially and use numbers for my worktree naming. So the main repo is always the left-most, then worktree 1, and so on. Rereading the Claude Code transcript is an excellent tool for getting back into the context of what I’m working on in a particular window. I struggle to run more than 3 truly complex implementations at once, but can handle as many as 6 or 7 if I’m just fixing little bugs.