Async AI Coding with LumifyDev
I use Claude Code every day. It’s the best coding tool I’ve used. But the one thing that always bugged me is that it’s synchronous. You sit there, watch it work, wait for it to finish. One session at a time.
As a solo dev, that’s a problem. I’d have five things I wanted to build but could only point Claude at one of them while the other four sat in my head. My bottleneck wasn’t writing code anymore — it was context-switching and waiting.
So I built LumifyDev, a CLI that connects LumifyHub kanban boards to async Claude Code sessions on a remote VM.
How it works
I add a card to a LumifyHub board with the task description. Then I run:
lumifydev run <card-id>
LumifyDev fetches the card’s title, description, and comments. It SSHs into my VM, creates an isolated git worktree, starts a tmux session, and launches Claude Code with the card context as the prompt. When the session finishes, it posts a comment back on the card with the session details.
I close my laptop and go do something else.
Later, I check on it:
lumifydev status <card-id>
That peeks at the last 30 lines of session output. If the work looks done, I pull the branch locally:
lumifydev checkout <card-id>
Review the diff, merge what’s good, send back what needs more work.
My actual workflow
Throughout the day, I add cards as tasks come to mind. End of day, I batch run two or three of them. Each one gets its own worktree so they don’t step on each other.
I check status from my phone using LumifyDev’s interactive mode — it’s a TUI with single-keypress navigation, designed for SSH from Termius on my iPhone.
Next morning, I pull the branches locally, review the diffs, merge. It genuinely feels like having a junior dev who works at 3am and doesn’t need standups.
The setup
The whole thing is a single Python script with zero dependencies. The VM side just needs SSH, git, tmux, and Claude Code installed.
I run mine on a Hetzner CPX11 — about $5/month. It handles two or three concurrent sessions without issues.
git clone https://github.com/LumifyHub-io/lumifydev.git
sudo ln -sf $(pwd)/lumifydev/lumifydev /usr/local/bin/lumifydev
lumifydev config
The config step walks you through connecting your LumifyHub workspace and VM. You need a LumifyHub API key from Settings and your VM’s SSH address.
What works well
The async loop is the real value. I spend less time watching Claude work and more time reviewing finished work. The worktree isolation means I can run multiple sessions without conflicts, and the card comments create a natural paper trail of what was dispatched and when.
The phone workflow is also surprisingly useful. I’ll be away from my desk and want to kick off a task or check on something — the TUI makes that possible without a laptop.
What doesn’t
The quality of the output depends entirely on how well you write the card description. Vague cards get vague results. I’ve learned to write cards the same way I’d write a prompt — specific about what I want, with enough context for Claude to make good decisions.
There’s no automatic code review or merging. You still review every diff yourself. For me that’s a feature, not a limitation — I don’t want code going in without me looking at it. But if you’re hoping for fully autonomous end-to-end, this isn’t that.
Links
- LumifyDev repo
- LumifyHub (free)
- Video walkthrough — full setup and workflow demo