How it works
You talk to one agent: the Orchestrator. It never does the domain work itself; every research question, PRD section, design decision, and line of code is routed to the right specialist. It maintains a project wiki, keeps a rolling progress ledger, and proposes stage transitions for your approval, so the project only advances when you say so. Each stage’s output becomes the next stage’s input. The PRD is written against your actual research; the screens are designed against your actual PRD; the code implements your actual screens on your locked design tokens.The stage map
QA and Deploy are not stages; they are on-demand actions you can trigger whenever there is something to test or ship. The QA Agent drives a real browser against your app and returns a verdict; the Deploy Agent takes the project from the sandbox to a live URL.
Stage transitions are proposed, not automatic. When a stage’s work is done, the Orchestrator proposes the move; you confirm it. Telling it to move on is consent to advance.
Where the work lives
Projects run in isolated cloud sandboxes with live preview URLs, so you can see the app as it is built. Generated code syncs to your own GitHub repo as it is written: full export, no lock-in. Your project context (PRD, research, designs) is also available in Cursor and Claude over MCP.Build vs Code vs Design
One account, credits shared across all three surfaces. Pick the one that matches how much of the pipeline you need.Build
The full pipeline. Use it when you want research, a PRD, designs, and a shipped product from one idea, with the whole trail documented.
Code
The AI app builder with a live sandbox. Use it when you already know what to build and want to go straight to working code.
Design
UI generation and templates. Use it when you need screens, brand systems, or design assets without the rest of the pipeline.
Agents
Meet the full roster: what each agent reads, produces, and how to steer it.