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

Per the Bolt pricing page, the 2026 free tier:

• 1,000,000 tokens/month soft cap; 300,000 tokens/day hard cap
• No rollover — free-tier tokens reset daily; only paid tokens roll over (up to one extra month)
• Bolt branding on deployed sites; public projects only

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

• Bolt and Replit are full-stack; v0 is UI-only. v0 feeds React components into an existing codebase — it won't ship a finished app 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 (~50K–80K tokens):

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.

The killer use case: pair bolt.diy with a free provider to escape the 1M-tokens/month ceiling without paying for Pro. Gemini (free tier, daily caps shown only in AI Studio) or Groq (1,000 requests/day) gives effectively unlimited personal use. Quality trails Claude Sonnet 4.6 on the hardest tasks, but for most CRUD and UI work the gap is small.

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, full access to Claude Sonnet 4.6. Restrictions: deployed sites carry Bolt branding, projects are public, 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?

Default is Claude Sonnet 4.6. Paid plans unlock Claude Opus 4.7 for harder reasoning. 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.

Related Reads

• Cline Free VS Code Agent: Which Models Finish the Job — the right next step after Bolt scaffolds your app
• Aider Free AI Coding CLI: Still Free, Now Stalled — terminal-first complement to Bolt's browser UI
• Free AI APIs in 2026: We Tested 10, Four Now Want a Card — pick a free model to pair with bolt.diy
• OpenRouter Free Tier Tested: 18 Free Models, One API Key — the easiest single key for bolt.diy or downstream Cline
• Supabase vs Neon: Free PostgreSQL Database Compared — the natural backend for Bolt.new apps