Vibe Coding Tools Compared: App Builders and Coding Agents

Vibe coding tools fall into two groups. App builders, such as Lovable, Bolt, v0 and Replit Agent, run in the browser: you describe an app, and they write the code, preview it, add a database and sign-in, and host it. Coding agents, such as Claude Code, work inside a code repository you already have, on your machine or in the cloud, with your tests and your review process. Pick a builder if you are not a developer or you want a working, hosted first version this week. Pick a coding agent if the code already exists, several people change it, or it has to last. Within the builders, the choice mostly comes down to stack and hosting: v0 for Next.js on Vercel, Bolt for JavaScript apps and Expo mobile apps, Lovable for full-stack web apps with its own backend, and Replit Agent for apps hosted on Replit with the widest range of outputs.

What vibe coding is, and how to do it without losing control of the project, is in what is vibe coding. This page is about choosing the tool.

Two kinds of vibe coding tool

• App builders own the whole path: editor, preview, database, authentication, hosting and a web address. Nothing to install, and the first working version comes quickly. The trade-off is that you work in their environment, on their stack, and move the code out when you outgrow it.
• Coding agents own none of that. They read and edit the code in a repository, run commands you allow, and hand changes back as commits or pull requests. Slower to start if you have no project yet; much better once you have one worth keeping.

Many projects use both in turn: a builder for the first version, then a coding agent in a repository once real users arrive. The handover, and where builders sit next to low-code tools, are in Lovable, Bolt and v0 vs Claude Code and vibe coding vs low-code.

The tools at a glance

Tool           Kind           Builds                         Code lives
Lovable        App builder    Full-stack web apps            Lovable; Git sync to
                              (Lovable Cloud backend)        GitHub, GitLab, Bitbucket
Bolt           App builder    Web apps, sites, Expo mobile   Browser tab; GitHub sync,
                              apps; JavaScript backends      zip download
v0             App builder    Next.js apps and interfaces    Vercel sandbox; GitHub
                              (React, Tailwind, shadcn/ui)   branches and pull requests
Replit Agent   App builder    Web and mobile apps,           Replit project; Git pane,
                              dashboards, AI tools           zip download
Claude Code    Coding agent   Anything in your repository    Your repository, from
                                                             the first line

Lovable

Lovable’s documentation↗ (opens in a new tab) describes it as “a full-stack AI development platform for building, iterating on, and deploying web applications using natural language.” Its built-in backend, Lovable Cloud, supplies the database, authentication, storage, edge functions and hosting, so a non-developer can get to a working app with accounts and data without setting up anything else. It focuses on web apps: SaaS and internal tools, dashboards, marketplaces and marketing sites. Code syncs to GitHub, GitLab or Bitbucket. Best for founders and product managers who want a real, hosted web app and a backend they do not have to assemble.

Bolt

Bolt, from StackBlitz, runs its whole development environment inside your browser tab, which is why it needs no setup. It builds websites, web apps and, through Expo, mobile apps. The limit is the language: Bolt’s supported technologies page↗ (opens in a new tab) says it “only supports JavaScript-based backends, so languages such as PHP or Python aren’t compatible.” Bolt Cloud adds a database, hosting and domains, and GitHub sync or a zip download gets the code out. Best for JavaScript web apps and for a first mobile app from the browser. More in Bolt.new.

v0

v0 is Vercel’s builder, strongest at turning a description, a screenshot or a Figma file into a polished interface and then growing it into an app. Its full-stack guide↗ (opens in a new tab) says it defaults to Next.js, and that other frameworks work but Next.js gives the most reliable results. Projects run in a Vercel sandbox and publish to Vercel. With GitHub connected, v0 commits to a working branch per chat and ships through pull requests, so your branch protection still applies. Best for designers and front-end teams already on Next.js and Vercel. More in v0 by Vercel.

Replit Agent

Replit Agent builds and hosts apps inside Replit. The Replit Agent overview↗ (opens in a new tab) says “No code or technical knowledge required,” and lists web apps, mobile apps, data dashboards and AI-powered tools among its outputs. It plans work as tasks on a board, tests apps in a real browser on paid modes, saves checkpoints you can roll back to, and keeps a production database it is not allowed to change. Hosting options include always-on and scheduled deployments, which suit bots and background jobs. Best for beginners who want everything in one place, and for apps that need to run continuously. More in Replit Agent.

Claude Code

Anthropic’s Claude Code overview↗ (opens in a new tab) calls it an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools, in the terminal, an IDE, a desktop app or the browser. It picks no stack and hosts nothing: it works with whatever your repository already uses, in any language, under rules you write in CLAUDE.md. That makes it the wrong first tool for someone who has never set up a project and the right one for anything already in a repository. Cursor, Codex, GitHub Copilot and Gemini CLI are the same kind of tool; best AI coding agents compares them.

Which vibe coding tool fits you

• A founder with no coding background testing an idea: Lovable or Replit Agent. Both include a database, sign-in and hosting, and neither asks you to install anything.
• A designer with a Figma file: v0, or Lovable, both of which import Figma designs.
• A product manager building a prototype to show users or engineers: any builder, with the plan kept outside the chat; see vibe coding for product managers.
• A first mobile app: Bolt through Expo, or Replit Agent, which lists mobile apps among its outputs.
• A Python or other non-JavaScript backend: not Bolt. Replit Agent or a coding agent in your own repository.
• A developer or team with an existing codebase, tests and CI: a coding agent such as Claude Code. A builder would mean moving the code into its environment.
• Anything holding customer data or payments: whichever tool wrote the first version, a repository you control and a person who reads the auth and payment code before launch. Vibe coding security has the checklist.

Free vibe coding tools

Most of these let you start without paying, with limits that change often. Lovable says you can start for free. Bolt has a free plan that publishes to a bolt.host address with daily and monthly usage limits. Replit Agent has a Free Mode with an included allowance, though browser testing needs its paid Power or Max modes. v0’s documentation does not name a free plan, so check its plans page. Claude Code is not included in the free Claude plan; it needs a Pro, Max, Team or Enterprise plan or a Claude Console account. Whatever you start on, check each vendor’s current plans before you rely on a limit, and stay on the plan you are on until the app is worth paying for.

What to check before you commit to one

1. Can you get the code out? All five can, but differently: two-way sync, one-way push, pull requests or a zip. Connect Git on day one.
2. Which languages and frameworks does it support? Bolt is JavaScript only; v0 works best in Next.js.
3. Where will it be hosted, and can you move it? Builders make their own hosting easy and yours possible.
4. What does undo cover? Restoring an earlier version often does not roll back the database.
5. Where does the list of work live? In every builder it is a chat history, which the next tool cannot read.

Keep the plan outside the tool

The work list is the one thing that should survive a change of tool. fenbs keeps it on a board that people and AI assistants share over MCP: each feature, enhancement or bug is a task with a note, a plan and a test status, moving through To Do, Next Up, In Progress and Completed, with History showing who changed each one. v0 and Bolt can add a custom MCP server by URL, and Claude Code connects in one command, so the same tasks follow the project from the builder to the coding agent. Choices you want every assistant to respect, such as the database you picked, go on the Decisions and rules page, which each connected assistant reads first. fenbs has no due dates or sprints; it is the shared list, not a project plan. The habits that keep an agent on track once you switch are in agentic coding best practices.