Bolt.new

Where Bolt sits

Bolt.new is a browser-based AI app builder from StackBlitz. You type what you want — "a marketplace for vintage camera gear with listings, search and Stripe checkout" — and Bolt generates a full-stack web application in the browser: frontend, backend logic, database, and a live preview you can click immediately. The code is fully editable in place, and when it is ready you deploy it to Netlify or export it. The technological trick underneath is StackBlitz's WebContainers, which run a real Node.js environment inside the browser tab — so the code Bolt writes is actually executed and previewed before you see it, not just rendered as text.

It belongs to the same category as Replit Agent and Lovable: prompt-to-app tools aimed at founders and builders rather than software engineers by trade. Where Replit leans platform (hosting, database, mobile apps) and Lovable leans design, Bolt's identity is full-stack depth in the browser and the ability to keep working in the code yourself. The current plan lineup at time of writing, per Bolt's pricing page: Free at $0 with 1M tokens a month and a 300K-token daily cap; Pro at $18/month billed annually (about $25 month-to-month per third-party tracking) starting at 10M tokens with rollover; Teams at $27/member/month billed annually; and Enterprise by custom contract.

What it actually does

The workflow is prompt → plan → generate → run → edit → deploy. Bolt writes the project, the WebContainer boots it, and the preview updates as it works. Error handling is part of the loop: when the dev server throws, the agent reads the error and attempts a repair without being asked. Your editing options are three: prompt again, edit the code directly in the built-in editor, or — the underrated part — upload a file (up to 100MB on paid plans) or connect MCPs and connectors for external context.

The token model is the economic core. Bolt meters AI usage in tokens, and its own help center is candid that most token consumption goes to syncing your project's file system to the model on each message. The larger the project, the more every message costs. Pro removes the daily cap, adds rollover, opens the newest AI models, custom domains, and removes Bolt branding; Teams adds centralized billing, admin controls, private NPM registry support and priority support; Enterprise adds SOC 2 Type II, SSO and unlimited tokens priced by consumption.

My testing notes

I built a two-sided booking app — services, availability calendar, booking flow — from one prompt. Bolt produced a working Vite/React frontend with a backend and a database in roughly two minutes of generation time, and the preview ran immediately. The first pass got the data model broadly right (services, slots, bookings with correct relations) and the UI wrong in the ways I now expect: generic styling, no empty states, one form without validation. Three follow-up prompts fixed those, each one cheap because the app was still small.

The project-size economics showed up on day two. Adding an admin dashboard to the same project made each subsequent prompt cost noticeably more tokens — consistent with Bolt's own explanation of filesystem syncing — and one loop where the agent failed to fix a state bug and retried cost me a visible chunk of the monthly allocation for no progress. My rule after that: for anything beyond a prototype, export to GitHub and continue in a real editor where my context window is free and my mistakes are cheaper.

The free tier deserves a fair description: 1M tokens a month with a 300K daily cap is enough to build and evaluate a small site, and hosting (including unlimited databases and up to 333K web requests a month) is included even at $0. It is one of the more honest free tiers in the category — a real sandbox, not a screenshot of one.

What it does not do

It does not scale to serious engineering work. Large codebases consume tokens per message, context quality degrades as the project grows, and the agent's ability to hold the whole system in mind weakens — the failure mode is a plausible-looking fix in the wrong layer. It is web-app focused: no native mobile output like Replit's mobile apps, no backend infrastructure beyond what the generated project contains. There is no team collaboration story at the free tier, and the token system means your real monthly cost is subscription plus whatever reloads finishing the project requires — a number nobody quotes upfront. And while the code exports cleanly to GitHub, Bolt is not where a production codebase should live long-term.

Practical build notes

A few things separate a good Bolt session from a frustrating one, and I learned them the expensive way. Prompt with structure: name the pages, the data model and the flows explicitly, and Bolt's first pass lands much closer than a vibe-shaped description. Budget tokens for the second act: my booking app's first build was cheap, every subsequent meaningful change cost more as the project grew, and the help center's own explanation (filesystem syncing per message) matches what I saw in the usage meter. And when the agent enters a fix loop — the same error, three plausible rewrites, no progress — stop it, export, and fix it yourself. In my testing that loop was where most wasted tokens accumulated, and no amount of patience made it cheaper.

The WebContainer substrate is genuinely different from a screenshot generator. The dev server runs in your tab, so what you see in the preview is the actual app with actual network requests — I verified a form submission hitting the generated backend and landing in the database before I ever deployed. That also means the browser is doing real work: on large projects, tab memory and CPU usage rise accordingly, and a low-powered laptop will feel it. It is the right trade for the category — real execution beats plausible rendering — but it is a real trade worth knowing about.

On the paid features I did not fully exercise: connectors and MCPs (Pro and up) are how Bolt reaches external systems, and the SEO tooling and custom domains are what turn a prototype into something a stranger can find and trust. Teams features — private NPM registries, design-system knowledge with per-package prompts, granular admin controls — suggest Bolt's growth path runs through small product teams standardizing their UI, which is a sensible place to compete against both Replit's platform and Lovable's design polish. Enterprise adds SOC 2 Type II and SSO, per the pricing page; I did not audit those certifications.

Verdict

Bolt is the fastest path from prompt to a running full-stack web app I have tested in the browser, and the free tier is generous enough to prove that yourself. I rate it 4.0 for core functionality that covers generate-run-edit-deploy end to end, value that is strong for prototypes and unpredictable for larger builds, usability that hides real infrastructure well, a growing connector and MCP ecosystem plus Netlify deployment, and reliability signals from StackBlitz's WebContainers track record — held to 4.0 by the token economics that make bigger projects progressively more expensive to work on.

Sources

• https://bolt.new/pricing
• https://support.bolt.new/account-and-subscription/tokens
• https://genvibe.pro/blog/bolt-pricing
• https://getcreatr.com/bolt-pricing-2026
• https://bolt.new