Quickstart
From a prompt to a running preview in about five minutes.
On this page
This walkthrough creates a small React application, asks the agent to change it, and opens a live preview. You need a verified account and a workspace where AI is configured (see the note at the end if it is not).
1. Create a project
- Open /new-project.
- Pick a template. For this walkthrough choose a React + Vite + TypeScript + Tailwind starter. Static HTML sites and Node + Express APIs are the other two families.
- Name the project and create it. The template's files are copied into your workspace; nothing is generated by AI yet.
2. Ask the agent for a change
Open the agent panel and describe what you want. A good first prompt is specific about the outcome and leaves the implementation to the agent:
Add a "Projects" page with a responsive card grid. Each card shows a title,
a one-line description and a status badge (draft / live). Use the existing
Tailwind setup and keep the current navigation.
Leave the mode on Build. The agent will:
- Load context (file tree, relevant files, your last messages).
- Propose a short plan.
- Edit files with its tools, streaming each change into the panel.
- Run a type-check and build inside the sandbox and repair errors (up to two attempts).
Every run starts with a snapshot, so you can review the per-file diff and revert the whole run if you do not like it. How the agent works.
3. Run the preview
Click Run. Codexa starts an isolated sandbox for the project, installs dependencies, runs the template's dev command and exposes it at a preview URL. Logs stream into the Logs panel. The viewport controls let you check phone, tablet and desktop widths.
If the preview does not start, the Logs panel shows the real error output from the sandbox; troubleshooting lists the common ones.
4. Edit by hand
Anything the agent wrote is just code. Open a file in the editor, change it, and press Ctrl/Cmd+S (autosave also runs on a short delay). Saved changes are synced into the running sandbox.
5. Deploy
Open the Deploy panel and choose Codexa hosting. The build runs in a sandbox and the output is served at a *.codexa.app style subdomain assigned by the operator's configuration, with history and one-click rollback. Deployment covers Vercel and custom domains.
If AI is not configured
The agent needs an Anthropic or OpenAI API key configured by the operator of your Codexa deployment. When none is present the agent panel shows an explicit AI is not configured state and the generate endpoint returns ai_not_configured. Everything else (templates, editor, previews, deployments) works without it.