Create a project on OriginAI

When you have a product idea, you can create a project on OriginAI, generate a first draft of the specs, refine it, and publish a release — all without writing any code first. This page walks through that flow end to end.

Before you start

You only need two things:

• A web browser.
• An OriginAI account. If you don't have one yet, you can start for free.

A clear product description makes the generated draft much better. It helps to know who the product is for, what problem it solves, a few of the key pages or flows, and any hard constraints. If you don't have all of that yet, don't worry — you can fill in the details as you refine the specs.

Try it now with an example

Not sure what to write? Click one of the examples below — it opens the OriginAI home page with the description already filled in. You can tweak it before generating, or just click Generate to see the full flow.

You can also copy a description manually and paste it into the composer at getoriginai.com, or open the link in a new tab and adjust the text before generating.

Create the project

There are two ways in, depending on whether you're signed in:

If you're not signed in yet, open getoriginai.com, describe your product in the composer on the home page, and generate. You'll then be asked to sign up or log in, and your description is normally carried through so generation can continue afterward.

If you're already signed in, open the app and either create a project directly from a description using Generate, or create an empty project first and write the specs yourself in the workbench.

Use whichever route is easiest for you right now — they produce the same kind of project.

How generation works: README first, then prototype

Generation happens in two distinct stages. Understanding this helps you know what to expect and when to intervene.

Stage 1 — Generate the product README

When the project is empty, the in-app agent starts by creating a README.rpml — a product-level document that captures the big picture:

• Who the product is for and what problem it solves
• The feature modules and how they relate
• The information architecture: the sitemap, navigation model, and which page is responsible for what
• A page/route plan with links to the screen files that will follow
• Key interaction flows (rendered as diagrams)
• Roles and permissions, when relevant

The agent deliberately does not create screen or prototype files at this stage. It focuses on getting the product structure right first, because everything else builds on top of it.

Once the README is generated, the agent tells you that you can start building the prototype. This is a good moment to review the README — is the information architecture right? Are the planned pages the ones you actually need? Fix the overview before generating screens, and the screens will be much better.

Stage 2 — Generate prototype screens from the README

When you're happy with the README (or when you click Start building prototype), the agent generates individual .rpml screen files — one per page or route from the README's plan. Each screen file captures:

• The layout and UI elements for that page
• The states it can be in (empty, loading, error, permission-restricted, etc.)
• Annotations and rules specific to that screen

The agent uses the README as the blueprint, so any structure you locked in during Stage 1 carries through automatically. If you later change the README's route plan, you can ask the agent to add or update screens to match.

The onboarding progress card

The workbench shows a progress card that tracks where you are in this flow:

1. Write the product README — the README.rpml exists.
2. Generate UI screens — at least one prototype .rpml file exists.
3. Select a region and leave a comment — you've used canvas pick + comment.
4. Publish a release — your specs are frozen into a shareable version.

You don't have to follow these in strict order, but the card helps you see what's left before you're ready to hand off.

Review the first draft

Once generation finishes, resist the urge to hand it off immediately. Spend a few minutes checking that the draft actually matches what you have in mind:

• Is the user right, and does the main flow make sense?
• Are any important pages missing, or are there pages you don't need?
• Does the canvas look like the product you're trying to build?

You can edit the specs directly, or select an element on the canvas and ask the in-app agent to adjust copy, structure, or flow. Remember that you're editing product specs, not application code.

When you're comfortable treating this draft as the source of truth for your product, you're ready to publish.

Publish a release

Publishing freezes your specs at a point in time so everyone works from exactly the same content.

1. Open the project's release panel.
2. Optionally add a note describing what this version represents.
3. Publish.
4. Keep the version details, and open the preview if one is available.

Your product will keep evolving, and that's expected. The loop is simple: edit in the workbench, publish a new release, and share the new release. That cycle is iteration in OriginAI.

Hand off the release

From the publish result you can usually copy a few different handoff options:

• A plain-text link for coding agents that can open a URL.
• An implement prompt for coding agents already connected to a repository.
• A preview link for people.

For details on which option to use when, see Share specs with your team or coding agent.

Troubleshooting

Related

• Already have code? See Create product specs from a codebase.
• Ready to hand off? See Share specs with your team or coding agent.