How to Build a Working Web App With Bolt.new in Under an Hour

November 8, 2026

Bolt.new is a browser-based AI builder that writes, runs and deploys full-stack apps from a prompt. You describe the app, it scaffolds React plus Node in a WebContainer, and you can ship a live URL the same session.

That sounds like magic. It isn’t. It’s fast, and it breaks in specific, predictable ways. This guide covers the setup path that actually works in 2026, what the free plan gives you, and the mistakes that burn tokens fastest.

What is Bolt.new and who is it for?

Bolt.new is an AI web app builder from StackBlitz. It runs a real Node.js environment inside your browser tab (WebContainer), so npm install and npm run dev happen locally without a server. You get a preview pane next to the chat, and a one-click deploy to Netlify.

Good fit:

  • Founders who need a clickable prototype for a pitch on Friday
  • Developers who want a throwaway tool: a CSV cleaner, a calculator, a landing page with a form
  • Anyone teaching a junior dev what a full-stack app looks like end to end

Poor fit:

  • Apps needing a database with row-level permissions and migrations you own (Supabase connections exist, but you’re one bad prompt from a broken schema)
  • Anything with heavy server-side compute or long-running jobs
  • Teams already deep in Next.js where a scaffold mismatch means rewriting half the repo

How do I write a prompt that doesn’t come back broken?

The first prompt decides 70% of the outcome. Bolt reads your whole prompt as a spec, so vague language gets filled in with its own defaults.

Use this order:

  1. App type and stack. “A React + Vite dashboard, Tailwind for styling, local mock data, no backend yet.”
  2. The one screen that matters. Describe the main view in plain layout terms: sidebar left, table center, filters top.
  3. Data shape. Give 3–5 field names and types. “Each row: name (string), status (one of ok/warn/fail), cost (number, USD).”
  4. Two interactions. Anything more and the first pass will be half-built.
  5. Constraints. “No external API calls. Keep everything in one src folder.”

A prompt that has worked well in practice:

Build a React + Vite app with Tailwind. One page: a table of API keys with columns for name, created date, and last used. Add a search box above the table and a button that opens a modal to create a new key. Use local state only, seed with 8 sample rows.

The modal, not the table, is where Bolt usually slips. Check the modal first, then the table filters.

What does Bolt.new cost and how fast do tokens run out?

Pricing as of late 2026: a free tier with daily token allowance (roughly 150k–200k tokens for a light day, reset every 24 hours), Pro at about $20/month, and Pro 200+ at higher tiers. Tokens are consumed per model response, and long files cost more than short ones.

Two practical limits bite people:

  • The free tier is fine for one small app. A 3-screen dashboard can eat a full day’s allowance before it works right.
  • Every fix costs tokens. Asking Bolt to “rebuild everything” re-generates whole files. Asking it to “change only the onclick handler in KeyTable.tsx” costs a fraction.

If you’re prototyping seriously, Pro is cheaper than the time you’ll lose waiting for a reset.

How do I fix the errors Bolt.new throws?

The console panel under the preview shows the real stack trace. Read it before you type anything into the chat.

Error you’ll see Usual cause What to ask for
Cannot find module Bolt installed the package but didn’t import it with the right path “Import X from the package we just installed and check the lockfile”
Blank white preview Runtime error on first render Open the console, paste the error text into chat
Infinite install loop Conflicting versions in package.json “Pin all dependencies to versions in the lockfile and reinstall”
Preview works, deploy fails Env vars missing on Netlify Set them in Netlify dashboard, not in chat

One trick that saves tokens: when the app is close to working, export it. Bolt’s “Export to GitHub” pushes the working repo out, and you can finish in your own editor with your own model credits. Half-built apps exported early are worth more than a perfect chat session you can’t reproduce.

When should you stop prompting and open the code?

Bolt has a code editor tab. Use it.

Signs the chat loop is failing:

  • The same error returns after two fix attempts
  • Bolt starts rewriting files you didn’t touch
  • The file tree grows without you asking for new features

When two of those happen, grab the code. A 5-line manual edit in the editor tab often beats a 4,000-token request. Developers who treat Bolt as a scaffold generator, not an oracle, get further.

Quick Summary

  • Bolt.new runs a full Node.js stack inside your browser via WebContainer; no local setup, no server needed until you deploy.
  • The first prompt should specify stack, one main screen, data fields and two interactions. Everything else is noise.
  • Free tier gives a daily token allowance; a single small dashboard can use a full day. Pro is roughly $20/month.
  • Fix small, ask for diff-sized changes, and export to GitHub the moment the app works.
  • Skip Bolt for anything requiring owned database migrations or long-running backend jobs.