Open any of these tools, type "build me a CRM," and ninety seconds later you have a running app. That is the demo, and the demo is identical across all four — which is exactly why it is useless for choosing between them. The first ten minutes are a solved problem. The question that decides whether you regret the choice arrives in week six, when the project has outgrown the prompt box and you need to know: can you take the code and leave?

That single axis — the exit — sorts these tools more cleanly than any feature list.

v0: the escape hatch

v0, Vercel's builder, is the one that answers the exit question with a shrug, because leaving was always the plan. It emits clean, idiomatic React and Next.js that you paste into a real repository and own outright. That design choice makes it the weakest of the four as a standalone "app builder" — it is happiest producing a component or a screen, not hosting your business — and the strongest as a generator you fold into work you already control. If you have an existing codebase and you want AI to write the tedious UI, v0 is the pick precisely because it doesn't try to keep you.

Bolt: fast because it's boxed

Bolt.new runs the entire stack inside your browser through WebContainers. That is genuinely clever and genuinely the source of both its strengths and its ceiling. The speed is real — there is no remote environment to provision, so iteration is instant, and it flexes across React, Vue, Svelte, and Astro with Expo for mobile. But a browser can only run JavaScript, so Bolt's backend is Node/Express, full stop. No Python, no Go, no PHP. For a prototype you intend to validate and then rebuild properly, that is a fine trade. Mistake it for a foundation and the wall arrives the first time you need a real backend.

The builders compete on how the app starts. You will choose, in the end, on how it ends — and three of the four make leaving harder than arriving.

Lovable: the path after the prototype

Lovable is built for what happens after the demo impresses someone: shipping an actual product. Its bet is the smooth path to the boring, necessary parts — auth, a database, payments — wired in through a tight Supabase integration, with onboarding gentle enough for non-engineers and account-based pricing that is cheaper for a team than Bolt's per-user token model. The trade is the usual one: the smoothness comes from building on Lovable's rails, and the further you go, the more leaving means rewiring rather than copying.

Replit: a whole environment, and a meter

Replit Agent is the only one of the four that is not really an "app builder" at all — it is a full cloud development environment with an agent driving it. Agent 3 can run autonomously for up to 200 minutes, spawn sub-agents for sub-tasks, roll back to checkpoints, and — the thing none of the others can do — stand up persistent server-side processes: a FastAPI service, a Go binary, a cron job, a webhook listener, a database migration it then verifies. If your app needs a backend that stays running, Replit is the answer here by default.

It also has the feature the others don't advertise: a meter. Replit moved to effort-based pricing, where each agent task bills by the compute and time it consumes — cents for a small edit, dollars for a feature. The Core plan's $25 of monthly credits sounds generous until an agent enters a debugging loop on a large project; users have documented single sessions burning $45 to $350. The autonomy that lets it run for 200 minutes is the same autonomy that lets it spend for 200 minutes. Budget for the runaway, not the happy path — the same lesson the rest of the agent economy keeps relearning.

The decision

Don't ask which writes the best code; on a clean prompt they are closer than the marketing admits. Ask which exit you want. Code you keep and fold into an existing repo: v0. A demo you'll throw away to test an idea: Bolt. A hosted product on rails, shipped fast with auth and payments: Lovable. A whole environment you'll live in, persistent backends and all: Replit — with one eye on the meter.

And once any of these has generated the app, something still has to read the diff before it ships. That's a separate tool and a separate decision — see CodeRabbit vs Greptile vs Qodo, and for the assistants that edit alongside you, Cursor vs Windsurf vs Copilot vs Claude Code.