Lovable is an AI app builder: you describe the product you want in chat, and it generates a working full-stack web app — typically React with Tailwind on the front end and Supabase for auth and data — that you can iterate on with follow-up prompts, then deploy and share from the same browser tab. It grew out of the open-source GPT Engineer project and has become one of the fastest-growing 'vibe coding' tools.
The trade-offs are the classic managed-platform ones: it's closed-source, you build on Lovable's credits and infrastructure, and the model choices are made for you. Heavy users hit the credit meter, and teams with data-residency or IP concerns often can't paste their product ideas into a hosted builder at all — which is exactly the gap the open-source alternatives below fill.
▶ See Lovable's homepage ↗The closest open-source match for the Lovable experience. Dyad is a desktop app: describe what you want, watch it scaffold a React app with live preview, and iterate in chat — but everything runs on your machine, with your own keys for OpenAI, Anthropic, Gemini or a local model. No credits, no lock-in, and the code is just a folder you own from the first prompt.
The official open-source version of Bolt.new, from StackBlitz's own labs. You get the same prompt-run-edit loop in the browser — the AI writes code, executes it in a WebContainer and shows you the running app — but you pick the model: OpenAI, Anthropic, Gemini, Groq or a local Ollama model. The most Lovable-like workflow you can fully self-host.
Onlook attacks the same problem from the design side: open a real React + Tailwind app, click any element and edit it visually — position, styles, copy — while Onlook writes the corresponding code changes. Paired with its AI chat, it's the pick for designers and founders who think in interfaces rather than prompts.
Where Lovable excels at prototypes, Pythagora (built on GPT Pilot) aims at complete products. It interviews you about requirements, plans the architecture, then writes, tests and debugs the app step by step — asking for input when it's unsure. Expect a slower, more deliberate loop that pays off on apps you intend to ship and maintain.
Not an app builder with a canvas — an autonomous software engineer. Give OpenHands a scoped ticket ('add Stripe checkout to this repo') and it works in a sandboxed environment: browsing docs, editing code, running the test suite and opening a PR. The pick when the app already exists and you want features built, not another prototype.
For Python teams, Reflex is the Lovable escape hatch: web apps in pure Python — no JavaScript — compiled to a React front end with a FastAPI-style backend. Its hosted Reflex Build adds the chat-to-app experience on top, and because the output is a standard open-source Reflex project, you can eject to your own infrastructure at any point.
Weights & Biases' answer to v0: describe a component and watch it render live, then ask for changes or convert the result to React, Svelte or plain HTML. It generates interfaces rather than full applications — but for landing pages, dashboards and design exploration it's fast, fun and entirely yours to run.
Frontman flips the Lovable workflow: instead of generating an app from scratch, it lives in your running app. Click any element in the browser, describe the change in plain English, and it edits the actual source files with hot reload — seeing the live DOM, component tree, computed CSS and server logs for context. Works with Next.js, Vite, Astro, Vue and Svelte.
A young but promising local-first builder: CodinIT installs as a desktop app on macOS, Windows or Linux and generates web and mobile apps from prompts — with your own model keys, entirely on your machine. The community is small and the edges are rough, but if you like Dyad's local philosophy and want mobile output too, it's one to watch.
The project Lovable itself grew out of. GPT Engineer is a CLI: write a prompt file describing the app, run one command, and it asks clarifying questions before generating the whole codebase. Development has slowed since the team went commercial, but it remains a clean, hackable foundation — and a piece of vibe-coding history.
The same 6-app build test for every tool: a dashboard, a CRUD app with auth, and a landing page — from first prompt to deployable code.
Commit velocity, release cadence and community health over the trailing 6 months — not just star counts.
How much of the Lovable workflow survives: chat-driven edits, full-stack scaffolding, deploy story and how easily you can export and own the code.
Popular proprietary tools people compare with Lovable — not open-source, but worth a look.
For most people, Dyad is the best overall — a free, local, open-source AI app builder that covers Lovable's chat-to-app workflow with your own model keys and no credit meter. bolt.diy is best if you want Bolt-style in-browser building, Onlook if you want to edit React apps visually, and Pythagora if you're building complete products rather than prototypes.
No. Lovable is a closed-source commercial product, though it originated from the open-source GPT Engineer project. If open source matters, Dyad, bolt.diy, Onlook, Pythagora, OpenHands, Reflex, OpenUI, Frontman, CodinIT and GPT Engineer are all genuinely open-source alternatives you can self-host.
Yes — every tool in this guide is free to self-host. Dyad and bolt.diy in particular are free to run locally; you only pay for the model usage on your own API keys, or nothing at all with a local model via Ollama.
Dyad, bolt.diy, Pythagora, OpenHands, Reflex and CodinIT generate full applications, including backend logic and auth. Onlook and Frontman focus on visually editing existing front ends, and OpenUI on generating UI components — pair those with a coding agent for the backend.
Dyad and bolt.diy both support local models through Ollama and similar runtimes, so you can build apps with zero data leaving your machine. OpenHands and GPT Engineer also accept any OpenAI-compatible endpoint, including local ones.
Notion is an all-in-one workspace that combines notes, docs, wikis, databases and lightweight project management in one flexible, block-based tool.
Claude Code is a terminal-based AI coding assistant built on Anthropic's Claude models.
n8n is a source-available workflow automation tool with a visual node editor, 400+ integrations and native AI/LLM nodes.
