Creating your first project

Choose a starting point, describe your product, and get an AI-generated project summary to build on. · Required

This is the entry point to the whole journey. It walks you from the dashboard to a reviewed project summary — the foundation every downstream generation builds on. If you'd rather begin from a ready-made foundation, see Starting from a template instead.

Step 1: Choose how to start

From the dashboard, click Create Your First Project — or the Create New Project card, once you have projects. VibeMap first asks How would you like to start? with two options:

• Start from an idea — describe your product and let the AI generate the full spec from your prompt. Best for new projects and early-stage ideas.
• Start from existing code — connect your IDE to VibeMap via the MCP server and let your AI coding agent scan your codebase and reverse-engineer it into structured specs. Best when a codebase already exists. Connecting over MCP is open on every plan, Free included; what your plan gates is which parts of the spec your agent may author — pages and schema need Starter or above. Full nuance in Two ways to generate.

You can use both flows later — they populate the same project. The rest of this page follows Start from an idea; for the code-first path see Connect your IDE.

Figure: Choose how to start.

Step 2: Fill in the project details

The Create New Project form has two fields.

Project name

A short, descriptive name — this is how the project appears in your dashboard and throughout the app.

• Keep it concise (2–5 words).
• Use the product name if you have one, or a descriptive label.
• Examples: "TaskFlow", "Recipe Sharing App", "Freelancer CRM".

Project description (prompt)

The most important field. Write a natural-language description of the product you want to build — the AI uses it as the foundation for everything it generates. The field arrives pre-filled with a template that prompts you to cover:

• What the app does — its core purpose and what users accomplish.
• Who it's for — the target users and their roles.
• Core features — the 3–5 main things users can do.
• What problems it solves — the pain points it addresses.
• Any specific requirements — stack, integrations, or unusual workflows (optional).

You don't need to be exhaustive — a few well-written paragraphs beat a vague essay. See Writing effective prompts for detailed guidance and worked examples.

💡 Tip: You don't pick a model here — VibeMap chooses one for you, and for a first summary that's a fast economy model, which is exactly what this step wants. On Pro and above a model selector appears next to Create Project, so you can send a complex or high-stakes generation to a stronger model.

Step 3: Submit and wait for generation

Click Create Project. The AI pipeline analyses your prompt and generates a project summary — a structured overview capturing your product's goals, target audience, value proposition, and scope. This usually takes 15–30 seconds; a progress indicator runs while it works. When it finishes, you're taken into the project.

Step 4: Take the first-visit tour

On your very first project, a short guided tour spotlights the layout you'll use everywhere:

1. Welcome — what VibeMap produces, and where the tour is about to take you.
2. Your project journey — the phase nav down the sidebar; the dots show what's already generated.
3. Always know the next step — the progress strip tracks the journey and points at the single best next action.
4. Meet your agent — the conversational agent can run generations, edit your spec, and answer questions; type a slash (/) to see every command, or /guide when you're unsure what's next. The tour introduces it on every plan, but the agent itself needs Starter and above.
5. Start with your first generation — generate personas first, then let features, stories, pages and schema build on top. Every generation is versioned, so nothing is final.
6. Where this leads — how the finished spec hands off to your own coding agent over MCP, or to the in-app App Builder.

The tour shows once, on your first project — it's tied to your account, not just this browser, so signing in on a new device or clearing site data won't replay it for an established user. Dismiss it any time with Skip or the ×, and reopen it whenever you like by adding ?tour=1 to any project URL.

Step 5: Review the project summary

The summary is the AI's interpretation of your idea, organised into clear sections — overview, target audience, core value proposition, key goals, and scope.

Figure: Project summary.

Review it carefully — it sets the direction for every downstream generation. If the AI misread something:

• Edit the summary directly to correct any misinterpretation.
• Regenerate it if the overall direction is off.
• Refine your prompt and start a new project if it's badly wrong.

Getting the summary right is worth the time: it cascades into better personas, features, and everything else. Full detail in Project summary.

Step 6: Generate the rest of the spec

With the summary confirmed, work through the journey. The sidebar and the progress strip recommend the order (green dots mark completed steps):

1. Business case — live-web-research market validation (Research phase; optional but recommended).
2. Personas — target user profiles.
3. Features — core product functionality.
4. User stories — requirements from the user's perspective.
5. Acceptance criteria — testable definitions of done (derived from features and stories).
6. Pages & sections — UI structure and routes.
7. Database schema — data model and relationships (pages first: page layouts give the schema generator better context).

Each step builds on the ones before it, so generating in order produces the most coherent result — but it's a loop, not a line: once pages exist you can cross-check acceptance criteria against them to close gaps. When the spec is complete, run Prepare for Dev to consolidate it into a build-ready blueprint.

Steps 2–5 run on any plan, Free included. Pages and schema (6–7) need Starter or above; the business case (1) and Prepare for Dev need Pro or above — see Account and billing.

Step 7: Hand off to your AI coding agent

Once the spec is in shape — features approved, acceptance criteria reviewed — connect your IDE's AI coding agent (Cursor, Claude Code, Windsurf, VS Code Copilot) to VibeMap over MCP. Your agent picks up work one acceptance criterion at a time, implements it in your codebase, and submits for review. Full walkthrough → Handoff overview.

⚡ Power-user hints

• Let the agent drive (Starter and above). Instead of clicking each generation, ask the conversational agent to run them ("generate personas, then features"). Its suggested-action chips always offer the right next step.
• Or let your own agent drive, on any plan. Connect an IDE agent over MCP and it can author the spec straight into the project on your tokens — see Two ways to generate.
• Don't skip the summary review. Two minutes fixing the summary saves regenerating every downstream artifact later.
• Experiment freely. Every generation is versioned — regenerate as often as you like and revert if a version was better.

↔ The traditional way

Kicking off a project by hand means a discovery workshop, a PM writing a brief, and a first PRD draft circulated for review — often a week before anyone agrees on scope. VibeMap compresses that into a single form and a 30-second generation, then keeps the resulting summary editable so you refine direction in place rather than in a doc thread.

What's next

• Writing effective prompts — sharpen your description for better output.
• Generating personas — the next step in the journey.