AI Web IDEs — Replit · Bolt.new · v0 · Lovable · Google AI Studio Build
AI Web IDEs — building apps from natural language, in a browser
If 10-ai-coding-ides covers desktop IDEs (Cursor · Windsurf · Kiro · Trae · Antigravity), this note covers their browser cousins — AI Web IDEs and AI app builders — that turn a prompt into code, DB, and deployment with no local install.
1. Categories
| Lane | Definition | Examples |
|---|---|---|
| AI Web IDE | Browser IDE + LLM (editing-led) | Replit · GitHub Codespaces + Copilot · StackBlitz |
| AI app builder (no-code) | Natural language → app (generation-led) | Replit Agent · Bolt.new · v0.dev · Lovable · Google AI Studio Build |
| AI coding IDE (desktop) | Local IDE + AI | Cursor · Windsurf · Kiro · Trae · Antigravity |
The lanes overlap. Replit is both IDE and Agent. Bolt.new leans app-builder but exposes an IDE.
2. Tool by tool
Replit (2016+, Agent 2024+)
- Browser IDE + multiplayer collab + deployment + Agent.
- Every language (Node · Python · Go · Rust · …).
- Full-stack: Reserved VM + Replit DB / PG.
- Models: Anthropic Claude (default), OpenAI / Gemini optional.
- Sweet spot: learning · pair programming · hackathons · Discord bots.
- Caveat: spiky traffic costs more than running on cloud directly.
- Deep dive.
Bolt.new (StackBlitz, 2024)
- WebContainer (in-browser Node VM) running Vite/Next/Astro generation.
- Site:
bolt.new. Built on StackBlitz tech. - Model: Anthropic Claude.
- Strength: 60-second SPA / Vite PoCs with instant preview.
- Weakness: limited long-running backend. Defers deploy to Vercel/Netlify.
- Fit: Vite/Next/Astro PoCs.
v0.dev (Vercel, 2023)
- UI-component-first. shadcn/ui + Tailwind.
- Site:
v0.dev. - Model: OpenAI GPT / Anthropic Claude (varies).
- Output: React/Vue/Svelte UI code (copy-paste) or Next.js auto-deploy.
- Strength: design-to-component. Prebuilt Tailwind + shadcn patterns.
- Weakness: no backend / DB automation. Frontend only.
- Fit: building UI parts of full-stack apps fast.
Lovable (2024, Sweden)
- Natural language → full-stack app + Supabase integration.
- Site:
lovable.dev. - Model: Anthropic Claude.
- GitHub auto-sync — exports code to your repo.
- Strength: full-stack (UI + DB + Auth) wired up via Supabase in one shot.
- Weakness: Supabase lock-in. Free quota is easy to hit.
- Fit: SaaS MVPs / learning full-stack apps.
Google AI Studio Build (Google, 2024+)
- Gemini-powered NL → app + Cloud Run auto-deploy.
- Site:
aistudio.google.com/apps. - Models: Gemini 1.5 / 2.0 / 2.5.
- DB: Firestore · Cloud SQL · direct Gemini calls.
- Strength: 1M-token context · Workspace integration · auto Cloud Run.
- Weakness: Google lock-in. Free key may train models.
- Fit: GCP users / long-context RAG / Workspace API combos.
- Deep dive.
GitHub Codespaces (2021+, GA 2022)
- Cloud dev containers usable from browser or VS Code desktop.
- Copilot integration for AI assistance.
- Free 60h/month (personal); org-level usage applies.
- Strength: turn any repo into a dev environment instantly. Enterprise.
- Weakness: AI is assistant, not protagonist. NL → app belongs more to Copilot Workspace.
StackBlitz (2017+)
- The original WebContainer — Node.js in the browser.
- Powers Bolt.new.
- Strength: 100% in-browser, sandboxed.
- Weakness: limited beyond WebContainer (no Python · Rust).
3. One-table comparison
| Tool | Launch | Model | Backend | Auto-DB | Deploy | Free tier |
|---|---|---|---|---|---|---|
| Replit Agent | 2016 / 2024 | Anthropic + opt | ✓ | Replit DB · PG | Replit | sleep free |
| Bolt.new | 2024 | Anthropic | partial | ✗ | Vercel/Netlify | limited |
| v0.dev | 2023 | OpenAI · Anthropic | ✗ | ✗ | Vercel | ✓ |
| Lovable | 2024 | Anthropic | ✓ | Supabase | Vercel/Netlify | ✓ |
| AI Studio Build | 2024 | Gemini | ✓ | Firestore · CloudSQL | Cloud Run | Gemini quota |
| Codespaces | 2021 | Copilot integration | dev container | user-defined | ✗ (CI) | 60h/mo |
4. Pick by scenario
| Scenario | Pick |
|---|---|
| Learning / pair programming / collab | Replit |
| Quick Vite/Next PoC | Bolt.new |
| UI components (shadcn / Tailwind) | v0.dev |
| Full-stack SaaS MVP + Supabase | Lovable |
| GCP / 1M-token context | AI Studio Build |
| Turn an existing repo into a dev env | Codespaces |
| Heavy-duty desktop IDE | Cursor / Windsurf (10) |
5. Data / training policy
Most free tiers may use input data for model training. For work code:
- Replit — Pro/Teams subscription + Privacy controls.
- Bolt.new — StackBlitz Pro.
- v0.dev — Vercel Pro.
- Lovable — paid plan + opt-out.
- AI Studio — paid key (Pay-as-you-go) or migrate to Vertex AI.
- Codespaces — enterprise policy (no training).
For work code, read the ToS and prefer enterprise plan / self-hosted.
6. Why benchmarks are hard
Same prompt, very different output across tools. Look at:
- Korean prompt handling (or your target language).
- Full-stack integration (UI + backend + DB).
- Code quality (modern React patterns · TypeScript · error handling).
- Deploy friction.
- GitHub sync (do you actually own the code?).
Try 30 minutes of your own scenario across each tool — the most accurate eval.
7. Shared limits
NL → app does not produce a finished SaaS:
- Complex business logic / domain models — written by humans.
- Auth · payments · email verification — extra wiring.
- Security (CSRF · XSS · SQLi) — not auto-applied.
- Region-specific integrations (e.g. Korean PG/AuthN) — extra work.
PoC → human polish → operate. That's the realistic loop.