← BLOG

How Spocket works

What actually happens between saying "deploy this" and your app being online. No marketing, just the sequence.

HOW IT WORKS · 23 JULY 2026 · 6 MIN READ

Spocket is hosting for things that have to keep running, driven from the AI editor you already build in. This page is the mechanical explanation: what happens, in order, and what is happening on our side while it does.

The one-time setup

You connect Spocket to your editor once. It uses OAuth rather than an API key, so a browser window opens, you approve it, and no secret is ever written into a config file you might commit or screenshot.

After that your assistant can see the Spocket tools and use them when the conversation calls for it. There is no CLI to install and nothing to keep updated.

What happens when you say "deploy this"

  1. 01
    Your editor reads the folderIt sends your source files and your manifest. Not node_modules, not .git, not your .env.
  2. 02
    We build a containerDependencies install from package.json or requirements.txt. Install scripts are disabled, so a package cannot run arbitrary code during the build.
  3. 03
    The process startsYour entry point runs in an isolated container with a hard memory cap. Nothing else on the machine can starve it.
  4. 04
    It stays runningNo idle timeout, no sleep, no cold start. If it exits, we start it again.

The whole sequence usually takes under a minute. The first deploy is the slowest because dependencies are downloaded fresh.

What happens when it crashes

It is restarted automatically. That is the normal case and it needs no attention from you.

What matters more is the abnormal case. If an app crashes repeatedly, roughly five times in half an hour, we stop retrying and tell you. A restart loop that runs forever looks like uptime on a dashboard while being completely broken, so we would rather show you a stopped app and a reason.

Where your secrets live

Environment variables are encrypted at rest and write-only. You can replace a value; you cannot read one back, and neither can a connected assistant. That is deliberate: a token nobody can retrieve is a token that cannot be exfiltrated by anything holding your credentials.

They are never printed in logs.

What your assistant is allowed to do

Deploying, reading logs, listing apps and setting variables happen freely, because they are routine and reversible. Restart and stop ask you first. Delete always asks.

Every deploy is kept, so a bad one is a rollback rather than an incident, and every action is recorded with who did it and when. The point is that an agent can do the ordinary work without you reviewing each step, and cannot do the irreversible work without you.

What it does not do

It does not host websites. If you built a static site or a Next.js app, Vercel and Netlify are better at that and mostly free. Spocket is for processes that run when nobody is looking: bots, workers, scrapers, scheduled jobs.

It also does not give you a machine to administer. There is no root, no SSH and no base image to pick. If you want that, a VPS is the right tool and costs about the same.

The panel

There is one, and you never have to open it. Logs, restart, environment variables and rollback are all there for when you want to click something rather than ask for it. Uploading a folder by hand works too, which is the path if you do not use an AI editor at all.

Need somewhere to put it?

Spocket runs apps, workers and scrapers that have to stay awake. Deploy from Claude or Cursor by asking, from $3/mo. First app is free for 7 days, no card.

Deploy an appRead the quickstart
Spocket
BlogDocsTermsPrivacyHome