Previous Show HNs for this project focused on the CLI/MCP side;
posting again because `backup drill` is the feature I think actually
solves a real problem people pretend they don't have.
Single Go binary (~15MB, zero deps) for homelab management. The backup drill feature creates an isolated container, restores your backup into it, and health-checks it. Your running services are never touched. Also handles Docker, WoL, network scanning, and multi-server SSH.
I run a few homelab servers and got tired of SSH-ing into each one every time I needed to check something. Especially at 3am when an alert fires — I just didn't want to open my laptop anymore.
So I built homebutler. It's a single Go binary (~15MB), zero dependencies. Point it at your servers via SSH and you get: system status, Docker control, Wake-on-LAN, port scanning, network discovery, alerts, and backup/restore. There's also a web dashboard (homebutler serve) and a TUI (homebutler watch).
It has a built-in MCP server so you can plug it into Claude Desktop, ChatGPT, Cursor, or whatever. But the AI only talks through structured JSON commands — it can restart a container but can't rm -rf anything. No raw shell access, ever.
I run a few homelab servers and got sick of opening SSH sessions
every time something broke at night. So I built this — a single
Go binary that handles status checks, Docker control, WoL, port
scanning, and alerts across multiple servers. It can also
self-heal basics like restarting crashed containers automatically.
The part I care about most: it has a built-in MCP server, so you
can plug it into Claude Desktop or whatever AI tool you use. But
the AI only talks to homebutler through structured JSON, it never
gets a raw shell. Felt important given how agents have been
behaving lately.
~15MB, zero dependencies, MIT licensed. Happy to talk about the
architecture if anyone's curious.
I run a few servers at home (Mac Mini M4 + Raspberry Pi) and got tired of the same 3 AM routine — SSH in, check what crashed, restart a container, go back to sleep.
I tried Portainer, btop, Uptime Kuma — all great tools, but I ended up with three dashboards open and still had to SSH in to actually fix things.
So I built homebutler. It's a single Go binary (~15MB) that packs a CLI, a TUI dashboard (Bubble Tea), and a web UI (compiled in via go:embed — no Node, no external assets). It manages Docker containers, monitors CPU/memory/disk across multiple servers over SSH, scans ports, sends alerts, and does Wake-on-LAN.
It also has a built-in MCP server, so AI tools can manage your homelab too — but it works perfectly fine without it.
The Game Boy Camera was 128x112 pixels with 4 shades of gray. The fact that people are still finding ways to pull images off these things almost 30 years later is peak hacker energy
I run a small homelab (Mac Mini + RPi5) and
tried Cockpit too. Great for single server
monitoring, but once I had multiple nodes,
I kept SSH-ing into each box anyway.
Ended up wanting something CLI-first that
could check all servers at once without
opening a browser. The web UI is nice for
a quick glance though.
Interesting approach using Signal for the transport layer. I've been working with real-time audio pipelines (chrome.tabCapture → Whisper) and the latency tradeoff between STT chunk size and accuracy is always tricky. What's the end-to-end latency like on a video call?
Interesting approach. The mmap streaming idea is clever, but I'd love to see real-world benchmarks beyond TinyLlama — especially for the 140B claim. Running that on a Mac Mini with 16GB would be the real proof point.
For context, I run a Mac Mini M4 as a homelab server and the memory pressure from even 7B models is noticeable. Curious how this handles sustained inference without thermal throttling.
Pretty lightweight — it SSHes into each box and runs homebutler locally there, so the overhead is basically one SSH connection + a quick read of /proc and the Docker socket per check. No background daemon sitting there polling, it only runs when you ask or when alerts --watch fires.
I run it across 3 machines (Mac Mini, Pi 5, and another box) and haven't noticed any impact. The binary itself is ~15MB and idles at zero CPU since it's not a long-running service.