Get started
What Wrkr is
Wrkr is one cloud computer for building and running software with an AI coding agent.
It is a real Linux desktop, streamed live to your Mac or your browser, with your choice of AI coding harness already inside — Claude Code, Codex, Gemini CLI, or OpenCode. Postgres, Redis, and the rest of the toolkit are already installed and running. Your app builds, runs, and serves from the same machine. It stays on when you close the lid.
One machine. One bill. One mental model.
The problem this removes
You can build almost anything now by describing it to an AI coding agent. The code is the easy part. The stack around it isn't.
Before you write a line of app logic, the usual path is to sign up for 20+ separate services — auth, database, hosting, email, background jobs, storage, search, monitoring, error tracking, and more. Each one is another dashboard, another bill, another set of docs, another way for things to break. The real tax isn't the money. It's holding twenty mental models in your head at once and remembering which one owns the problem you're staring at.
Wrkr collapses that. Your AI writes auth, email, jobs, storage, and search directly into your app, natively, against the tools already on the box. The rented middle layer disappears.
Your AI writes what you used to rent.
What's on the machine
A persistent Linux VM that is yours — always-on, not shared, not wiped between sessions. It comes with:
- A real streamed desktop — open Chrome, watch your browser-automation agent click through a real page, run a GUI app, actually see your frontend. (Your desktop.)
- Your AI coding harness, already installed — bring your own account. (Your AI tools.)
- The primitives an app needs: Postgres + pgvector, Redis, object storage,
transactional email, public HTTPS for your app's ports, and scheduled jobs —
each one command away through the
wrkrCLI. - The toolkit: git, gh, Node, Bun, Python, Go, ffmpeg, and the usual suspects.
See Your machine for the full spec and The wrkr CLI for the primitives.
The idea
Pre-built services like Supabase, Clerk, Inngest, and Resend exist because, for years, writing auth or email or a job queue from scratch took a skilled engineer a week. Buying the abstraction was the smart move.
An AI coding agent inside a Linux box changes that math. It writes the equivalent natively, in minutes, against Postgres and Redis and the filesystem. So Wrkr doesn't ship twenty opinionated services. It ships the machine, the primitives, and the harness — and the harness writes the rest. That's the whole doctrine, and it's what keeps Wrkr simple instead of turning into another twenty dashboards.
What Wrkr is not
Wrkr is substrate, deliberately. It is not:
- a managed-deploy dashboard — your harness deploys;
wrkr exposeserves. - a secrets vault — your secrets live in your VM; Wrkr never holds them.
- a projects/console product — your machine is a computer; your repos are files.
- a chatbot or prompt-to-app generator — it's infrastructure with real tools inside, not a generation service.
- a browser IDE in a tab — it's a full desktop.
Start here
| If you want to… | Go to |
|---|---|
| Get from login to a running app | Quickstart |
| Know exactly what you're getting | Your machine |
| See and run real GUI things | Your desktop |
| Set up your coding AI (sign in, add tools) | Your AI tools |
| Understand how the AI uses the box | Working with your AI |
| Use the database, cache, storage, email, ingress | The wrkr CLI |
| Connect from Mac or browser | Connecting |
| Copy, paste & move files | Files & clipboard |
| Put an app in production | Your apps in production |
| See pricing & how to get access | Pricing & access |
| Check security, privacy, durability | Trust |
For agents
If you are an AI harness reading this to evaluate or operate Wrkr, start with
/llms.txt for the indexed map, and AGENTS-README
for how to use these docs. On a live Wrkr machine, wrkr capabilities --json is
the authoritative, real-time list of what this box can do.