Bolt.new

AI app builder (browser) · by StackBlitz · official site

What it actually does

Bolt.new is a browser-based AI app builder from StackBlitz that takes natural-language prompts and generates full-stack web application code in real time. It runs entirely inside a browser-based Web Container (Node.js + npm sandbox), so you see a live preview of your app as the AI iterates. You can edit generated code directly, re-prompt to modify features, and deploy the app to a public URL with one click (via StackBlitz's hosting or Netlify/Vercel). The underlying model is a fine-tuned LLM (details vary) trained on StackBlitz's runtime environment. In practice, it produces code in React, Next.js, Express, and other popular frameworks.

Who it's for

Solo founders and early-stage product teams who need a functional prototype in under two hours and don't want to touch a terminal. Also for non-developer builders (product managers, designers) who want to validate an idea by seeing a working UI connected to a backend. If you're a senior engineer who values fine-grained control, Bolt.new will feel like a toy — but that toy can spit out a passable MVP faster than you can scaffold a project.

What works

Sub-minute first prototype. Describe "a Task Manager with CRUD operations, user authentication, and a dashboard" and Bolt.new generates a working app structure (Frontend + Express API + SQLite) in 20-30 seconds. The live preview updates as you prompt.
Accurate scaffolding for common stacks. Like v0.dev, Bolt.new knows the entire project structure: package.json, route handlers, React components, Tailwind classes. It avoids the "spaghetti file" trap many AI code generators fall into.
Seamless dependency resolution. The Web Container resolves npm packages in real time, so npm install happens automatically. No "missing module" errors if the AI chooses a package that exists.
Deployment is genuinely one-click. After generation, you can deploy to a generated .bolt subdomain or connect your own Netlify/Vercel account. The deployment pipeline is UI-based and doesn't require you to touch a CLI.
Iteration speed for small UI changes. Changing a button color or rearranging a layout is fast: type "make the sidebar collapsible" and the AI rewrites the relevant component in seconds, preserving the rest of the app state.

What breaks

Authentication implementation often fails. Bolt.new will write the frontend Login form and a Passport.js route, but the actual sign-in flow usually breaks on the first try — token storage, session middleware, or password hashing can be miswired. Debugging this in the browser-only console is painful because the AI doesn't persist context across iterations unless you manually re-explain.
Complex state management. The AI defaults to local React state (useState) even for global things. Trying to add Redux, Zustand, or TanStack Query often results in inconsistent code: providers not wrapped, selectors referencing missing slices. Rewriting that manually takes longer than starting from scratch.
Third-party API integration is hit-or-miss. Prompts like "add Stripe checkout" generate a plausible function, but the AI skips webhook handling, error callbacks, and idempotency keys. You need to fill those in yourself, and the environment stops short of giving you Stripe test keys or a dashboard to test them.
The environment is surprisingly fragile for database persistence. SQLite (the default) lives in memory between prompts. If you refresh the browser tab, the data is gone. Bolt.new doesn't persist a database file between sessions unless you explicitly deploy to a persistent backend. For a "working app" test, this means you can't reload the page and expect to see your data.
Deployment limits on the free plan. The free tier allows only one public deployment and limits compute hours per month (varies). Once you deploy, any further prompt iterations create a new project — you lose versioning. The paid plans (Bolt Pro, $20-50/month) increase limits but still throttle API calls.

Pricing reality

Free tier is generous for exploration: ~20 prompt runs per day, one deployed app at a time. Bolt Pro ($20/month) raises the daily limit to ~200 prompts and allows 5 concurrent deployments. Bolt Team ($50/month) adds team collaboration, custom domains, and higher API rate limits. If you exceed prompt limits, you cannot generate code until the next day — no pay-as-you-go token option as of 2026. For a single 1-hour test session, the free tier is sufficient, but you might hit the prompt limit if you do many iterations. If you need to build and deploy an app with a real data backend (e.g., PostgreSQL), you'll need a paid plan to use external services and persistent storage.

The honest comparison

v0.dev (Vercel): Vastly superior for UI-first projects — the generated Tailwind CSS is cleaner, animations are smoother, and the design system integration is tighter. But v0.dev does not generate backend code. If your app needs any server logic (auth, APIs, background jobs), Bolt.new wins by default. v0.dev also still requires a local dev environment for full-stack tasks. Replit AI (Ghostwriter): More mature for iterative editing within an existing codebase. Replit's AI understands project context better and can refactor across files without losing state. The tradeoff is that you must write at least some code yourself; Bolt.new offers a higher starting abstraction (prompt → full app). Replit also provides a persistent filesystem and database (built-in Postgres). For the "1 hour working app" test, Replit is more reliable for data persistence but slower to generate the initial scaffold. Lovable.dev (formerly GPT Engineer): Similar to Bolt.new but with a stronger focus on React + Supabase stacks. Lovable.dev produces more production-ready backend patterns (row-level security, migrations) out of the box. Bolt.new's advantage is that it is browser-only (no install) and slightly faster at code generation. Lovable.dev tends to generate code that is harder to edit manually due to higher abstraction. Verdict for the 1-hour test: Bolt.new is the fastest way to get a full-stack app skeleton running in your browser, but it will fail you if the app requires real user authentication, persistent data across sessions, or any third-party API that needs secret keys. You can ship a working CRUD demo in 1 hour — just don't reload the page and ignore the broken login flow.

When to use it

Use Bolt.new when you need a visually interactive prototype of a full-stack idea in under an hour and you don't mind that the code underneath is a scaffolding that needs manual hardening before it can face actual users.

Last verified: 2026-06-08 by kernel.