Overview
Bolt.new is what happens when you take an AI coding assistant and force it to actually do the work instead of just giving you advice. Created by StackBlitz, it is a browser-based, full-stack development environment that allows you to prompt a complete web application into existence.
The core problem with using standard LLMs like ChatGPT for coding is the "copy-paste loop." You get a snippet, paste it into VS Code, encounter an error, paste the error back, and repeat. Bolt removes that friction. It combines a chat interface with a fully functional IDE. When you ask for a "SaaS dashboard with a Supabase backend," Bolt generates the file structure, writes the React/Next.js code, installs the npm packages, and runs the app instantly in a preview window.
It is particularly popular with two distinct groups: solo founders who want to ship MVPs in a single afternoon, and non-technical folks (PMs or designers) who want to build internal tools without fighting a local development environment.
Key Features
1. WebContainers (The Engine Room)
This is the technical backbone that separates Bolt from generic chatbots. It uses StackBlitz's WebContainer technology to create a full Node.js runtime directly in your browser. This means the code runs on your local machine's CPU, not on a remote server. The result is zero latency when saving files or seeing updates. You can access a real terminal, run scripts, and install dependencies just like you would on a local machine, but without the "it works on my machine" headaches.
2. Full-Stack Context & "Plan Mode"
Most AI tools differ on frontend vs. backend, but Bolt handles both simultaneously. It can configure database connections (like Supabase), handle API routes, and manage state. A standout feature here is Plan Mode. Before writing a single line of code for a complex request, Bolt generates a structural architectural plan. You can review this logic first, which saves a massive amount of tokens and frustration by preventing the AI from coding itself into a corner.
3. Smart Diffing & Error Correction
In the early days of AI coding, asking for a small change often resulted in the AI rewriting the entire file. Bolt has introduced smarter "diffing" capabilities. It attempts to lock specific files and only applies changes where necessary to preserve existing functionality. If the AI generates code that throws a terminal error, it can often read the error log itself and apply a fix without you needing to copy-paste the stack trace.
4. Bolt.diy (Open Source Flexibility)
For developers wary of platform lock-in or high costs, the team released Bolt.diy. This is an open-source version of the engine that supports a "Bring Your Own Key" (BYOK) model. You can hook it up to standard providers like OpenAI and Anthropic, or use Ollama to run local models. This allows power users to bypass the token markups of the SaaS platform while keeping the same interface.
Pricing
Bolt uses a token-based system. Tokens are burned when the AI reads your codebase or generates new text.
- Free Tier: ~1 million tokens/month (with a daily cap of roughly 300k). Good for testing the waters or very small scripts. Includes standard hosting.
- Pro ($20/mo): 10 million tokens/month. This is the entry point for serious usage.
- Pro 50 ($50/mo): 26 million tokens/month. Better for active builders iterating on a launch.
- Pro 100 / 200 ($100-$200/mo): Scales up to 55M and 120M tokens respectively. Intended for power users or agencies shipping client work.
Important Note: As of late 2025, paid tokens roll over for one month (60-day expiry), which helps if you have a quiet month followed by a sprint.
Pros & Cons
The Good
- Zero Setup: You go from a blank tab to a running Next.js app in about 30 seconds. No
npm installwait times or environment variable hell. - Context Retention: It manages multiple files surprisingly well. It knows that changing a component in
/uirequires an update to the page in/app. - Deployment: One-click deployment to Netlify or Vercel makes the "idea to URL" pipeline incredibly short.
- Multi-Model Access: You can toggle between Claude (Haiku/Sonnet/Opus) and GPT-4o depending on whether you need speed or complex reasoning.
The Bad
- The Token Burn: Despite improvements, the AI can still be "over-eager." Sometimes it will rewrite a 500-line file just to change a variable name, which drains your token allowance rapidly.
- The Complexity Wall: Bolt is fantastic for the first 80% of an app. However, once a project gets massive, the AI can get confused, start looping errors, or introduce regressions. It is not yet a replacement for a senior engineer on a legacy codebase.
- Support Latency: Users have noted that getting help with billing or stuck projects can take longer than expected.
Verdict
Bolt.new is currently the best-in-class tool for greenfield projects and rapid prototyping. If you are an indie hacker wanting to launch a SaaS, or a product manager wanting to validate an idea without waiting for engineering resources, this tool is worth the subscription.
However, if you are working on a massive, existing enterprise application, you might find the token limits and context windows restrictive. For those users, it serves better as a sandbox to test new features in isolation before porting them to the main repo.
Recommendation: Start with the Free Tier to understand the workflow. If you like it, jump to the $20 Pro plan, but keep an eye on your token usage. If you are a developer comfortable with API keys, look into the Bolt.diy repo to run it on your own terms.