Bolt.new: Free AI App Builder That Runs in Your Browser

On this page
  1. The WebContainer Trick
  2. Free Tier: What 1M Tokens Buys
  3. Bolt.new vs Lovable vs v0 vs Replit Agent vs Cursor
  4. Five-Minute Quickstart
  5. Making the Free Tier Last
  6. When Bolt Hits Its Limits
  7. Bolt.diy: The Self-Hosted Alternative
  8. Frequently Asked Questions
  9. Verdict
Quick answer: Bolt.new is a free, browser-based AI app builder from StackBlitz: you prompt it, it scaffolds a full-stack web app, runs it inside your browser tab via WebContainers, and one-click deploys to Netlify. The free tier gives 1,000,000 tokens/month with a 300,000-token daily cap — realistically 3–8 meaningful prompts a day. To escape the ceiling, scaffold in Bolt then continue in a free local CLI agent, or self-host the open-source bolt.diy with your own free API key.

You type a prompt — "build a Next.js todo app with Supabase auth" — and Bolt scaffolds the project, runs npm install, boots a dev server, and shows a live preview, all in a browser tab. The twist: the Node runtime, package manager, and file system all run inside your browser via StackBlitz WebContainers, so edits are instant with zero cold start. By 2026 it's one of the three most-used AI app builders alongside Lovable and v0.

The WebContainer Trick

Most AI coding tools either run in your local editor (Cursor, Cline, Aider) and assume your toolchain is set up, or run on a remote cloud VM (Replit Agent, Codespaces). Bolt is a third camp: the Linux runtime, package manager, file system, and HTTP server are compiled to WebAssembly and run in a single browser tab. Open Bolt on a fresh Chromebook and have a running app in 60 seconds. Concretely:

  • Zero cold start — the dev server is ready the moment the agent finishes writing files.
  • Compute is your laptop, not their cloud — StackBlitz pays nothing per running project, which is why a $25/mo Pro plan can bundle 10M tokens.
  • Instant fork/remix — sharing a Bolt URL means anyone who opens it gets a running clone in seconds.

The limit: WebContainers can't run native binaries not compiled for WASM — no Docker-in-Docker, native Postgres, or NumPy/Pandas data science (some Python works via Pyodide, but slow and limited). For React, Next.js, Vue, Svelte, Astro, Vite, Tailwind, and TypeScript prototyping, none of that matters. For ML pipelines or heavy native deps, Bolt is the wrong tool.

Free Tier: What 1M Tokens Buys

What we checked, and what we didn't (2026-10-11). We do not have a Bolt account, so we have not run a prompt or measured token use ourselves: the per-prompt token figures and timings on this page are estimates from Bolt's docs and third-party write-ups, not our measurements. What we did re-check on 11 October 2026 is every plan detail against Bolt's pricing page and help center, plus the competitors' pricing pages. That re-read corrected five things on this page:

This page used to sayThe vendor's page says (2026-10-11)
Free projects are public onlyFree includes "Public and private projects"
Default model Claude Sonnet 4.6; paid plans unlock Opus 4.7Bolt now offers a "Standard" and a "Max" agent and picks the model itself; free gets Standard only. No model is named.
Free tokens reset dailyThe 300K cap resets daily; the 1M allowance resets on the 1st of each month. Free tokens never roll over.
v0: chat-only free tier, UI components only, $20/monthv0 Free can deploy apps to Vercel, with a 7-messages-a-day limit; the paid tier is Plus at $30/user/month
Replit Agent had only an expiring trial; Cursor free = 2-week Pro trialReplit lists a free Starter plan with daily Agent credits; Cursor lists a Free plan with limited Agent and Chat usage

The current free tier, per Bolt's pricing page:

  • 1,000,000 tokens/month soft cap; 300,000 tokens/day hard cap
  • No rollover — the daily cap resets every day and the monthly 1M on the 1st; only paid tokens roll over (up to one extra month)
  • Bolt branding on deployed sites; public and private projects are both allowed

How much is 300K tokens? Less than you'd expect. Each prompt resends the entire file tree, conversation history, and tool definitions to Claude. A medium edit to a 5-file Next.js project ("add a search bar that filters the post list") typically burns 30,000–60,000 tokens. An independent 2026 comparison pegs realistic output at "roughly 3–8 meaningful prompts per day." Enough to scaffold a small project and refine it — not enough to ship a real app in one sitting.

Two workarounds: (1) Plan offline first — write a tight design doc before opening Bolt; throwaway prompting destroys the free tier. (2) Use Bolt for scaffold, then export — the "Download" button ships a full-project zip; open it in Cline or Aider with a free Gemini or OpenRouter key and iterate with no token pressure.

Bolt.new vs Lovable vs v0 vs Replit Agent vs Cursor

FeatureBolt.newLovablev0Replit AgentCursor
SurfaceBrowser (in-browser runtime)Browser (remote sandbox)Browser (UI-only)Browser (remote VM)Local IDE (forked VS Code)
Free tier1M tokens/mo, 300K/day5 build credits/day (max 30/month)7 messages/day, can deploy to VercelStarter: free daily Agent credits, limited intelligenceFree plan: limited Agent and Chat usage
Stack scopeFull-stack web (Node)Full-stack webUI-first; can deploy full appsFull-stack + DB + cronAnything on disk
Default modelNot named ("Standard" agent on free)Claude / GPT (managed)v0-managedClaude / GPT (managed)Cursor-managed
One-click deployNetlify (built in)Built-inVercelReplit DeploymentsExternal
DatabaseConnect SupabaseConnect SupabaseSupabase / NeonBuilt-in PostgresWhatever you wire
Open-source clonebolt.diyNoNoNoNo
Paid pricing$25/mo (10M tokens)$25/mo$30/mo (Plus)Check current Core billingPlus $8/mo; Premium $30/mo

The free-plan entries above were checked against v0 pricing, Replit pricing, and Cursor's current plan documentation on 2026-10-24. Cursor's older Hobby plan is now listed in its legacy documentation; the current plan is named Free.

  • Bolt and Replit are full-stack; v0 is UI-first. v0 is strongest at React UI and can now deploy apps to Vercel, but it is built around components rather than a full backend with auth and a database.
  • Bolt's WebContainer beats Replit's sandbox for iteration speed. Replit is more capable on long-running backend tasks (real Linux VM), but every file edit waits on the cloud sandbox; Bolt feels instant because the runtime is local.
  • Lovable targets non-developers who want polish over control. Want to read and edit the code yourself? Bolt is far more developer-friendly.

Five-Minute Quickstart

Shortest path from nothing to a deployed Next.js app on the free tier:

  1. Sign in at bolt.new (any Chromium browser; Firefox works, slightly slower). GitHub or email, no credit card.
  2. Write a tight first prompt. Vague prompts force follow-ups that each burn context; a specific one produces a working scaffold in a single pass.
  3. Run the preview — mounts as soon as npm install finishes (~8–15s).
  4. Connect Supabase (optional) — click "Connect Supabase" to auto-create a free Supabase project and inject credentials, then prompt Bolt to migrate off JSON (~40K tokens).
  5. Deploy → Netlify → ~20s → a real public URL. Custom domains and removing Bolt branding are paid.
  6. Push to GitHub — creates a public repo; clone locally and continue in Cline/Aider without paying Bolt's token budget.

A first prompt specific enough to produce a runnable app in one pass (roughly 50K–80K tokens — an estimate, not a measured run):

Build a Next.js 14 (App Router) blog with:

Pages:
- /            list of posts (title, date, excerpt, tag pill)
- /posts/[slug] full post with markdown rendering
- /admin       password-gated form to create posts (env var ADMIN_PASSWORD)

Data:
- Posts in a single posts.json file
- Each post: slug, title, date (ISO), tag, excerpt, body (markdown)

Styling:
- Tailwind CSS, dark mode default, monospace headlines

Tooling:
- TypeScript strict mode
- next.config.js with images.unoptimized = true (deploys cleanly to Netlify)

After scaffolding, add three sample posts.

Making the Free Tier Last

The 300K daily cap is the real constraint. Five techniques that extend it:

  • Front-load detail. One specific 2,000-token prompt beats five 500-token clarifiers — each follow-up resends the whole history, so cost grows quadratically with chattiness.
  • Pin a tight system prompt. Project-level rules ("Always use TypeScript. Never install packages without asking.") are sent once per request and stop Bolt re-deciding conventions each edit.
  • Edit small files yourself. Bolt's editor is real — fix a typo or Tailwind class by hand instead of spending 20K tokens.
  • Use Discussion mode for architecture talk — smaller context, no file tree or diff machinery.
  • Export and continue locally once you hit the cap — download the zip, open in Cline + a free OpenRouter or Gemini key.

When Bolt Hits Its Limits

  • Native binary dependencies — FFmpeg, Postgres-the-binary, Python with NumPy/PyTorch, Rust. WebContainers run a Node-shaped runtime; calls into system libraries mostly don't work.
  • Long-running background workers — a WebContainer lives only while the tab is open. No daemon, cron, or queue worker. Replit Agent's real VM fits better here.
  • Large monorepos — Bolt is happiest with up to ~a dozen workspaces; a 500-package pnpm monorepo tanks the UX. Use Cline or Aider locally.

For landing pages, internal admin tools, prototypes, indie SaaS MVPs, and hackathon submissions, Bolt is best-in-class.

Bolt.diy: The Self-Hosted Alternative

StackBlitz's open-source bolt.diy ships the same chat-and-preview UI but lets you bring your own model and run it on a free hosting platform.

FeatureBolt.new (hosted)bolt.diy (self-hosted)
CostFree (1M tokens/mo limit)Free + your own API costs
ModelManaged by Bolt (Standard / Max agents, model not named)Any model — bring your own key
Free model optionNo (always burns Bolt tokens)Yes — Gemini or Groq, subject to their limits
One-click deployBuilt-in NetlifyManual
Setup time0 minutes~30 minutes
PrivacyCode goes through Bolt serversStays on your machine

The practical use case: pair bolt.diy with a provider that offers a free API tier, such as Gemini or Groq. You avoid Bolt's token meter, but the provider's own rate and usage limits still apply. Check those limits before a large project.

Together AI is supported as a bring-your-own-key provider, but it is not a free-model option: Together's detailed billing docs require a $15 first credit purchase. Later top-ups have a $5 minimum.

git clone https://github.com/stackblitz-labs/bolt.diy
cd bolt.diy
pnpm install

# bring your own key in .env.local
echo "GOOGLE_GENERATIVE_AI_API_KEY=your_gemini_key_here" > .env.local
# or: echo "GROQ_API_KEY=your_groq_key_here" >> .env.local

pnpm run dev   # open http://localhost:5173

Frequently Asked Questions

Is Bolt.new really free?

Yes. 1,000,000 tokens/month with a 300,000 daily cap, no credit card, on Bolt's "Standard" agent. Restrictions: deployed sites carry Bolt branding, no custom domain, and you can't lift the daily cap without upgrading. Genuinely usable for prototyping and learning; sustained daily work will hit the cap.

Which AI model does Bolt.new use?

Bolt no longer names one. As of October 2026 it offers two agents, Standard and Max, and "handles model selection behind the scenes"; the free plan gets Standard only. You pick the depth/cost trade-off, not the provider.

Can I use my own API key?

Not on hosted bolt.new — the model is managed and bundled into the token cost. For BYOK use the open-source bolt.diy fork, which plugs in Gemini, Groq, OpenRouter, Together AI, or local Ollama.

How does the token meter count?

Tokens reflect what Claude charges StackBlitz: input (prompt + file context + history) plus output (the diff written back). Per the official token docs, file context dominates — touching a 1,000-line file costs roughly the same regardless of prompt length. That's why "edit only the relevant files" is a real technique.

Verdict

Bolt.new is the best answer to "I have an idea and want a running app in 5 minutes." The WebContainer trick is a generation ahead of tools that wrap a remote VM in chat — no cold start, no setup. The 300K daily cap is real and follow-up prompts add up fast, so the right free-tier workflow is: front-load detail, get a runnable scaffold in one pass, push to GitHub, and continue refinements in Cline or Aider against any free API. To escape the ceiling entirely, run bolt.diy with Gemini or Groq.