Build Something in Five Minutes
The quickest way to see what Circuitry does is to ask for something and watch it appear: one sentence in the ask box, and a working page opens beside the chat.
What you'll do
Ask for a small game, then play it. A coding agent writes the files, opens the result in a browser tab next to the chat, and — if something doesn't work — clicks through the page itself to find out why.
1. Open a project
Choose File → New Project and pick a folder. This is where the agent writes its files, so what it builds is real work on your disk rather than something trapped in a chat window. The project name appears in the corner of the Welcome tab.
In the desktop app this just works. On iPhone, iPad, Android or in a web browser, the folder lives on a computer you have paired with Circuit, or on a cloud machine — that is where builds and coding agents actually run. Circuitry reaches it from whichever device you happen to be on.
2. Ask
Type into the ask box on the Welcome tab:
Build a side-scrolling game where I tap to fly through gaps in pipes, then open it in the browser so I can play.
With a project open, a build like this is the coding agent's job: the chat opens with your words already in the message. Press send.
If you have not set an agent up yet, the ask box says so and offers every way forward: a subscription with cloud AI, a paired computer with a coding agent on it, a cloud computer, or your own models. Pick one and come back — your words are waiting in the chat.
3. Choose an agent
The first time you send, a short setup asks which agent you want. There is no wrong answer — you can change it later — but the choices fall into four groups.
- Circuitry cloud AI — included with a subscription. Nothing to install or configure; it works from every device.
- Free to start — Gemini CLI signs in with a personal Google account and includes a free allowance; no card needed.
- A subscription you already have — Claude Code or Codex CLI. These appear when the tool is installed on the computer running the project, and use the plan you already pay for.
- Your own key, or your own machine — OpenRouter takes one key and gives you a model picker covering many providers, including free options. Ollama runs models privately on your own computer and also offers cloud models; LM Studio runs them locally; on a Mac, MLX models are set up and run by Circuit, with nothing to install yourself. Qwen Code works with your own key or a local model.
Each CLI agent wants its own sign-in the first time: Gemini CLI, Claude Code and Codex ask you to log in, and Qwen Code asks which provider to use. That happens inside the CLI itself. Press the terminal button in the chat header ("Run … in a terminal") to open the agent as a terminal and finish its setup there; it keeps full access to your Circuitry documents in that mode.
4. Watch
The agent will:
- Write the files — an HTML page, and whatever script and styles it needs.
- Open it in a browser tab right beside the chat, so you see the result immediately.
- Play it, if you asked it to — clicking and pressing keys on the page to check the game actually runs.
5. Play it, then ask for changes
The page is live and yours. Play it, then say what you want different, in the same chat:
Too hard — make the gaps bigger and slow it down a bit.
Add a score in the corner and a game-over screen.
The agent edits the files and reloads the tab. You keep playing.
What just happened
Three things worth knowing, because they apply to everything else you'll do here.
The result opened where you can see it. Circuitry has its own browser tab, so a page the agent builds lands in your workspace beside the work it came from — not in a window somewhere else that you have to go and find. The tab stays open afterwards.
The agent can use that tab too. It isn't just a viewer. The agent can click buttons, type into fields, scroll, read what the page printed to its console, and see which network requests failed. That's how it checks its own work instead of asking you whether it worked. See Testing in the browser.
You can watch, and stop it. While the agent is using a page, the edges of that tab glow, a pointer moves to each thing it clicks, and a bar names what it's doing — with a Stop button. Nothing happens out of sight.
Where to go next
- Testing in the browser — everything the agent can do with a page, including checking phone layouts.
- Your first workflow — when you want something to run repeatedly, on a schedule or from a webhook, rather than once in a chat.
- The coding agent — plan approval, diffs, and working with your own repository.
- The smart whiteboard — when it's easier to draw what you mean than describe it.