#A throwaway server, from a shell

A server bought for an hour, enrolled with Lore, deployed to, tested and deleted, with every step a command and no web page in the loop. This is the loop an agent runs to try a change on a real machine.

It was walked on 2026-11-06 on a Scaleway DEV1-S (about €0.009 an hour, Ubuntu 24.04): about ten minutes from buying the server to deleting it.

It needs the Lore CLI and Bay of the release after 0.35.0, which added the commands below.

You need the Lore CLI signed in (lore login), a provider CLI (here Scaleway's scw, set up with an SSH key), jq, and a Lore project to work in:

lore project create scratch
lore project capability enable apps --option stacks

capability enable keeps the options already on (Deploy apps) and adds the ones named.

#1. Buy the server

scw instance server create type=DEV1-S image=ubuntu_noble name=scratch tags.0=throwaway -o json | jq -r .id > server-id
scw instance server wait "$(cat server-id)"
IP=$(scw instance server get "$(cat server-id)" -o json | jq -r .public_ip.address)
BASE=${IP//./-}.sslip.io   # 203-0-113-10.sslip.io

The server is billed by the hour from here on. Its IP gives it a name with no DNS to set up: <anything>.<ip-with-dashes>.sslip.io resolves to it.

#2. Enrol it with Lore, in one line

SECRET=$(lore estate add scratch-vps --allow-deploys --ttl 2h --output json | jq -r .secret)
ssh root@$IP "curl -sSL https://alepha.dev/bay/install.sh | LORE_URL=https://lore.alepha.dev LORE_ESTATE_SECRET=$SECRET BAY_BASE_DOMAIN=$BASE sh"
lore estate wait scratch-vps

• --output json hands the one-time secret to the script; --ttl 2h marks the estate overdue in two hours and tells you once, so a forgotten server does not go unnoticed. Lore never stops or deletes it.
• With LORE_URL and LORE_ESTATE_SECRET, the installer installs Docker and Bay, writes and starts the systemd unit, waits for bay serve, and runs bay connector set. The secret is never printed.
• BAY_BASE_DOMAIN is what apps are named under (<app>.<base>); without one, Bay refuses to deploy an app.
• lore estate wait returns once the machine is online. lore estate get scratch-vps then shows the public IP Lore saw it connect from.

On the machine, bay status, bay doctor and bay connector show need no flags: they find the installer's /opt/bay/data and /run/bay/control.sock.

#3. Deploy a stack, on HTTPS

1# compose.yml2services:3  web:4    image: traefik/whoami:v1.105    x-lore:6      port: 80

lore stack create --file compose.yml --estate scratch-vps
lore stack deploy -e staging --test-hostname web
curl https://staging-scratch.$BASE/

--test-hostname web routes <env>-<stack>.<ip>.sslip.io to the web service before the deploy, and Bay obtains a Let's Encrypt certificate for it on the first request: on the walk, the first curl already answered over a verified certificate.

#4. Deploy an app

An app scaffolded for Lore deploys with one command, which ends with where it answers:

npm create alepha hello -- --infra lore --lore-project scratch
cd hello
npx alepha deploy
# → https://hello.203-0-113-10.sslip.io

--infra lore writes lore({ project }) into alepha.config.ts and mounts the sigil, so the app reports its page views, Web Vitals and errors to the project from the first deploy. lore deploy -t latest -e production redeploys a stored build the same way.

#5. Delete everything

lore stack destroy -e staging --yes
lore stack delete --yes
lore apps instance delete --app hello --env production --yes
lore estate delete scratch-vps --yes
scw instance server terminate "$(cat server-id)" with-ip=true with-block=true
lore project delete scratch --confirm scratch

Order matters, and Lore says so: an estate a copy or a stack still deploys to is refused (naming them), and so is a stack with an environment left. estate delete revokes the machine's secret but stops nothing on it, which is why the server goes with its provider, with its IP and its disk: a plain delete leaves both billing. scw instance server list, scw instance ip list and scw block volume list come back empty afterwards.