Get started
Quickstart
This is the whole path, start to finish: sign in, your machine opens, you build something with your AI agent, and you put it online. About ten minutes, most of which is the agent writing code while you watch.
You need nothing installed. A browser is enough to start. (For the daily-driver experience, install the Mac app — but that can wait.)
1. Sign in and open your machine
Wrkr is in invite-only beta, so you sign in with the credentials you were given. Go to wrkr.dev and log in.
Your machine wakes and connects. You'll see it move through Connecting → Opening machine → Live, and then a real Linux desktop resolves in front of you. This is your computer — the same desktop, the same files, every time you come back.
Take a second to notice what's there: a terminal, a file manager, Chrome, a text editor. It's a normal Linux desktop, because it is one.
2. Start your AI coding agent
Open the terminal and start the harness you want. All four are already installed:
claude # Claude Code
codex # Codex
gemini # Gemini CLI
opencode # OpenCode
The first time, the harness will ask you to log in to your own account for that tool. Wrkr doesn't bundle or bill AI usage — you bring your own harness account and your own model API keys. (Why: Working with your AI.)
Your machine already knows how to help. Wrkr seeds each harness with steering so that when it needs a database or email or file storage, it reaches for what's already on the box instead of signing you up for another vendor.
3. Build something
Tell the agent what you want. For a first run, keep it real but small:
"Create a new project in
~/notes. A single-page app with a Postgres-backed list of notes I can add to and delete. Use the database that's already on this machine."
The agent will scaffold the project, write the code, and — because of the steering — ask the box for a database instead of reaching for a hosted one. Under the hood that's:
wrkr db # prints a ready-to-use Postgres connection URL
One command, no signup, no keys. Postgres (with pgvector for embeddings) is
already running; wrkr db just hands your app its URL. The same is true for
cache, object storage, and email.
4. Run it
Let the agent start the app, or run it yourself. Say your app is listening on port 3000:
# whatever your project uses — e.g.
bun run dev
Because this is a real desktop, you can open Chrome right on the machine and
load http://localhost:3000 to see it. No "tests are green but the UI is broken"
guessing — you're looking at the actual thing.
5. Put it online
One command publishes a running port at a public HTTPS URL:
wrkr expose --port 3000
You get back a live https://…apps.wrkr.dev URL with TLS already handled. Share
it. It stays up because your machine stays on — even after you close the browser
and walk away.
Want it on your own domain? Add --domain:
wrkr expose --port 3000 --domain app.example.com
wrkr expose prints the DNS record to add at your registrar, and serves your app
there once it resolves. Details: Ingress & custom domains.
What just happened
In ten minutes you signed in to a persistent cloud computer, built an app with an AI agent, gave it a real database with one command, saw it running in a real browser, and served it to the internet — on one machine, with one bill. No Supabase, no Vercel, no Resend, no dashboards.
Next
- Your machine — exactly what you're running on.
- The wrkr CLI — database, cache, storage, email, ingress, cron, backup.
- Working with your AI — get the most out of the harness.
- Your apps in production — keep it running and put it on a domain.