daybox.dev

A beefy devbox that exists only while you're using it.

One CLI summons an hourly-billed machine in your cloud account, drops you in over your private net, and reaps it when you walk away — while your repos, your auth, and your tmux session survive on a persistent volume. The next box comes up already knowing who you are.

$ curl -fsSL https://daybox.dev/install.sh | sh

then daybox init once, daybox up forever · macOS & Linux · single static binary · read the source · the installer verifies the release against a hash pinned in the script — read it first, it's short

what you get

Lifecycle on demand

daybox up boots a fresh box from your provider's API; daybox down — or the idle reaper, ~30 min after you leave — deletes it. It bills only while it exists.

Memory that survives

Repos, agent auth, git identity, tmux — all on a persistent /work volume. The machine is disposable; your workspace isn't.

Dark to the internet

Each box joins your private WireGuard mesh (tsnet + self-hosted Headscale) on first boot. No public listeners — ingress from your net, structurally, not as a firewall posture to audit.

Plain ssh, no vendor app

The mesh node lives inside the CLI: no root, no TUN, coexists with a corporate VPN. A one-line ProxyCommand makes vanilla ssh, scp, rsync, and VS Code Remote just work.

Your cloud, your bill

You bring a scoped token; your provider bills you directly at raw-VPS prices. We never resell compute. Hetzner today, AWS-with-spot next.

Fails closed

The installer pins the sha256 of the release's checksum manifest in the script itself — a hostile artifact store can't alter or roll back a release — and daybox init verifies a pinned minisign signature. Pin mismatch, bad signature, checksum mismatch — it aborts. There is no advisory mode.

how it works

Two boxes: a small always-on control plane — a cheap VPS, or any box you already leave on — that holds your cloud token and runs the reaper and your private mesh; and the devbox, a beefy hourly machine that exists only while you're in it.

  1. Install — the one-liner above drops a single static daybox binary on your PATH.
  2. daybox init — interviews you for a project-scoped Hetzner token, provisions a small control-plane VPS (or adopts one you already have) running the coordination server and the reaper, enrolls your laptop on your net, and writes the config for you. It shows you the live price before creating anything. Idempotent — re-run it any time to heal drift.
  3. daybox up — the control plane summons the big box, attaches your /work volume, and hands it a single-use ephemeral key to join your mesh. You ssh in. When you stop, it's reaped and vanishes from daybox net — no ghosts.

the cost math

thingbilling
control plane (always-on VPS)flat monthlyfrom ~€5.49/mo
big box (ccx33 · 8 vCPU / 32 GB)hourly, only while it exists~€0.23/h
50 GB /work volumealways~€2.86/mo

prices are Hetzner's, from their API at deploy time (stamped 2026-08-15; big box priced in the default location, control plane varies by location — the CLI shows the live number before creating anything).

The reaper deletes a box ~30 min after your last session (it won't kill a detached agent mid-task), force-reaps after 1 h unreachable, and a hard lifetime cap (default 12 h) catches runaways that always look busy. Nothing zombie-bills overnight. daybox status always shows age, spend so far, and time remaining.

faq

Why not just Tailscale?

It's built on the same tech (tsnet + WireGuard). Tailscale gets you into a box; it doesn't create, bill, or reap one, and it doesn't remember your workspace. daybox owns the lifecycle + memory; the mesh is a layer it embeds, not competes with.

Why not Codespaces?

Codespaces runs on GitHub's compute and bill, scoped to a repo, environment rebuilt not remembered. daybox runs in your account, any repo, one persistent workspace, and bursts to a bigger machine at raw-VPS prices.

Why not Coder?

Coder is a great team platform, but it's a server to run and Terraform to write. daybox is one binary + a small always-on VPS, personal-first, zero-dependency.

Why not fly.io / devbox SaaS?

They resell you compute — their margin, their abuse problem. daybox is BYO cloud; your provider bills you and I never touch your compute plane.

curl | sh, really?

The installer is served over TLS from daybox.dev and pins the hash of the release it attests; every download is checksum-verified against that anchor before installing (plus a minisign signature when you have minisign), so a compromised artifact host can't tamper — or serve you an old release. You can read it first — it's short — or skip it entirely: every release ships a signed source tarball with a documented reproducible build, so you can build the exact released binary yourself.

Where's the source?

The code is MIT and lives in the open at github.com/daybox-dev/dayboxgit clone it and cmd/daybox/build.sh builds the binary you'd otherwise install. Every release also ships a minisign-signed SHA256SUMS and a source tarball at daybox.dev/dl, and the build is reproducible: rebuild the tarball and you get, bit for bit, the binary you installed — what you read is what runs (see VERIFY.md).

Isn't a persistent volume with plaintext creds a risk?

Yes, and it's in the security doc as an accepted v1 limitation. Seed only scoped, revocable, spend-capped keys. The encrypted per-session bundle is the planned fix.

Does it lock me to Hetzner?

No — Hetzner is just the reference provider. The abstraction is five primitives (summon / reap / probe / attach-volume / user_data), deliberately not baking in Hetzner-isms. AWS-with-spot is the next target (and spot fits perfectly: state's on the volume, the reaper already treats a vanished box as survivable).

Business model?

It's a free, open-source personal tool and stays that way. If a hosted team tier (paid coordination, BYO compute) ever happens it's an upgrade of something already loved, not the point.