Lovable

The choice depends on what you are bringing: an idea without code, your own project, or business plans.

Who it is for. For people with an idea and no code: an entrepreneur with a service concept, a marketer with a landing page, a designer with a mockup. Describe it in words, get a working prototype with a database and password login, and show it to people. The free plan is enough to get a feel for it - five credits a day.

Who it is not for. If a project already lives in its own code, this is a job for Cursor, Claude Code and Codex: they work in your repository. Lovable builds only from scratch in its sandbox and does not import outside repositories. The genre handles complex logic poorly: in an independent comparison of builders, only blogs and storefronts worked reliably.

Launching a business on this? Then someone who can read code is mandatory between "built" and "released." Apps in this genre have common vulnerabilities, and Lovable's have already led to leaks of other people's data. Section 6 explains exactly what to check.

Just getting started? If you have not built projects yet, start with the "What is vibecoding" page and come back here for a tool.

What vibecoding isStarting from scratch: how to give AI tasks with no coding experience

If it fits, start with section 3. If you are choosing between builders, see the fork with its neighbor in section 7.

The full cycle lives in one tab: from a phrase to a published app. Step by step.

Builds from words. Type what you need into the "Ask Lovable to create..." chat field and watch a live site appear beside it. You can see the first version in a few minutes. Practitioners have built working apps in three to four hours (review, 23.08.2026).

Thinks like an agent. Since June 2025, an agent mode has operated inside: the platform plans, reads logs, finds and fixes errors on its own instead of waiting for a command at every step. The vendor claims 90% fewer build errors - its own figure, without external measurement.

Handles the database and login. Since fall 2025, there has been a built-in Supabase-based Lovable Cloud backend: tables, user registration, and files. You do not need to build a separate server for a prototype.

Adds AI to your app. Ask it to "translate reviews and summarize them," and Lovable AI connects Gemini models without your keys or separate billing.

Publishes it itself. The Publish button deploys the project to built-in hosting. You can connect a custom domain on paid plans and buy one right inside.

Exports code. The project syncs both ways with GitHub, but only through one branch, and an outside finished repository cannot come in: Lovable builds from scratch. You can download the code as an archive on paid plans.

Integrates into other chats. Since 15.07.2026, a published app can be offered as an MCP server: users can call it directly from ChatGPT or Claude.

In our project. An honest note: this site is built by coding agents Claude Code and Codex; Lovable is not part of our pipeline. This card was assembled from fresh research, with every figure checked against primary sources on August 23.

There is nothing to install; Lovable lives in the browser. Three steps to your first prototype.

From Russia. The site lovable.dev is accessible, and registration works without workarounds, according to Russian reviews as of 23.08.2026. There is one barrier, payment: Stripe does not accept cards from Russian banks, so for a paid plan an international card is required.

Everything is in the browser. The builder, database, and hosting live on the site, while builds run on Lovable's servers. No powerful hardware is needed.

Try it free. Free gives five credits a day, up to thirty a month. Enough to build and explore a first version, not enough for serious work.

Step 1. Go to lovable.dev and sign in with email or Google - no card is needed, and the free plan starts immediately.

Step 2. Describe the app in chat; Russian is fine. The more specific your request about screens and data, the fewer credits will burn on rework.

Step 3. Watch the live preview, request edits, and hit Publish when the version is ready to show people.

Builders are not measured by agent benchmarks. Lovable is measured by the market, and the market has two stories about it.

Record growth. One hundred million dollars in annual revenue in its first eight months; by June 2026, the run rate had reached five hundred million. Both figures came from the company itself; the press only reported them.

Money believes. In August 2026, investors put in another 400 million dollars at a 13.3 billion valuation, with Tencent joining the round. Its genre neighbor Bolt last reported revenue of 40 million, in spring 2025.

Traffic cooled off. Barclays analysts counted Lovable traffic down 40% by September 2025 from its peak. Paying users remain and pay more: revenue retention is above 100%. Whether this is the genre's peak or a pause, the market has not decided yet.

Self-reported figures. 60 million projects and “two thirds of the Fortune 500” - figures from the vendor itself, with no methodology. We report them with that caveat.

Plans are simple; the catch is usage: you pay with credits, but you only find out how many credits a request uses afterward.

Plans. Free gives five credits a day (no more than 30 a month) for $0. Pro starts at $25 for 100 credits a month; with annual billing, the same credits cost $21. Business starts at $50 and adds enterprise SSO with a security center.

A credit is charged per turn. A chat message costs about one credit, a complex task in agent mode costs more, and a small edit costs less. Hence the main practical equation: the more precise the prompt, the cheaper the app.

The main practitioner complaint. Credits burn on fixing the builder’s own errors: in one breakdown, 80% of credits went to repeated requests and fixes, while removing an ad banner cost 10–12 credits. The remedy is unchanged - targeted edits and a specific request.

What a real app costs. Reviews in the category cite a Reddit case: a 29,000-line app cost its author $250. The order of magnitude is worth remembering - a substantial project costs hundreds of dollars in credits, not the subscription price.

From Russia. The subscription is paid by foreign card through Stripe; Russian cards do not work - the path is described in section 3.

Three honest questions: whose code, where prompts go, and how many holes the platform has.

Your code is private. Projects are private by default on all plans. Code leaves through GitHub sync, or as an archive on paid plans.

Where prompts go. Generation is performed by external Google and OpenAI models; Lovable has no model of its own. That means build quality, but also dependency: prices and limits are ultimately set by other labs.

First wave: phishing. In April 2025, Guardio Labs researchers showed that Lovable was the easiest in the category to persuade into building a phishing page and hosting it itself. In their test, Lovable scored 1.8 out of 10 for security; ChatGPT, by comparison - 8.

Second wave: open databases. In June 2025, it emerged that generated apps were shipping with databases open for anyone to read and write: vulnerability CVE-2025-48757 was confirmed in 170 live apps. Lovable fixed the generation of new schemas within a week, while owners had to lock down old apps manually.

Third wave: the platform itself. In early 2026, a researcher showed that other people’s source code, keys, and AI chats were accessible from a free account. The company did not address the report for 48 days and initially said there had been no breach. The real count in this incident: 18,697 student records from one app, including 4,538 minors.

What this means. Since April 2025, publishing includes a built-in security check, but it does not replace a human reading the code. If you build something on Lovable with other people’s data, show the project to someone who understands access rights.

The category’s main duel. Both build an app from a description in words and both start at $25; from there, their paths diverge.

Where Lovable is stronger. The path is smoother: fewer buttons, less code in sight, backend and AI features are enabled from chat. The scale is its advantage too - 60 million claimed projects and a $13.3 billion valuation.

Where the rival is stronger. With Bolt code is visible and editable as you work, builds run in the browser, and mobile preview comes through Expo. And it alone has an open fallback, bolt.diy: the same mechanics with any models.

Models. Lovable offers Gemini and GPT to choose from, with Gemini 3.7 Flash as the default. Bolt uses Anthropic models only.

Looking beyond the category? Nearby v0 and Replit, and from Russia, with ruble payments - Russian Lork.

The product is under two years old, yet its history already includes record growth, two rounds, and the archive of its progenitor. Key dates below.